DEV Community: Tarun yadav

MeloTTS on Mac: Install It Yourself or Use an App

Tarun yadav — Mon, 09 Nov 2026 17:55:25 +0000

MeloTTS is an MIT-licensed multilingual TTS library with no Mac app. See the real install steps, the MPS gap, and when a packaged voice app is simpler.

Short answer: MeloTTS is an MIT-licensed multilingual text to speech library from MIT and MyShell.ai that its docs describe as fast enough for real-time inference on CPU, and it can be made to work on an Apple Silicon Mac through its Python package. The project does not ship a Mac app, does not document macOS as a tested target, and Murmur does not run MeloTTS. Your honest options are to install the official repository yourself, or to use a packaged Mac app with a model it actually ships, such as Kokoro.

Search results for MeloTTS mix several different artifacts: the GitHub repository, the Hugging Face model cards, a hosted demo, and third-party wrappers. Only the repository and cards describe what the project supports. This guide sticks to those, plus the public issue tracker for the macOS failures users have already reported, and it separates what the project claims from what a Mac user has to do to get audio out.

What MeloTTS actually is

The README calls MeloTTS a high-quality multilingual text to speech library. It is a VITS-family model, and the acknowledgements say the implementation is based on Coqui TTS, VITS, VITS2, and Bert-VITS2. That lineage is acoustic-model work rather than a large language model that predicts speech tokens, though the project publishes no parameter count. The package exposes three entry points: a Python API, a command line tool, and a small Gradio web interface.

The repository is version 0.1.2 according to setup.py, and the install documentation says the project is developed and tested on Ubuntu 20.04 with Python 3.9. That sentence is the most important one for a Mac reader. Nothing in the README promises a supported macOS build, and the same install page tells macOS users who hit problems to try Docker instead. You can still run it natively, but you are using an untested path.

Two claims in the README are worth keeping separate. First, the project supports English, Spanish, French, Chinese, Japanese, and Korean, with five English accents. Second, the model is described as fast enough for CPU real-time inference. That second claim is the project's own statement about its intended deployment, not a benchmark on your machine, and it does not say which CPU or which sentence length was used. Measure it on your own Mac before you plan a long narration session around it.

Languages, accents, and speakers the README lists

MeloTTS uses one checkpoint per language, and the speaker set inside a checkpoint is fixed. The README and the install guide name these speaker IDs:

Language	Speaker IDs named in the docs	Notes from the README
English	EN-Default, EN-US, EN-BR, EN_INDIA, EN-AU	Five accents. The install guide lists them as the available English speakers.
Spanish	ES	One speaker in the example.
French	FR	One speaker in the example.
Chinese	ZH	The README says the Chinese speaker supports mixed Chinese and English text.
Japanese	JP	One speaker in the example.
Korean	KR	One speaker in the example.

Notice what is not in that table: no German, Italian, Portuguese, or Arabic, and no per-language list of dozens of voices. The English checkpoint gives you five accents, but the docs list one speaker ID per accent, not a catalogue of men, women, and ages. If your script needs several distinct English speakers in conversation, you are changing pitch and speed on the same voice or training something of your own.

The Chinese mixed-language behavior is a real differentiator and it is stated plainly in the README. If your script switches between Chinese and English inside a sentence, that is the case MeloTTS was built to handle. Do not assume the same mixing works for Japanese or Korean, because the README does not claim it.

The official install path on macOS

These are the install commands from the project's own install document, unchanged except for leaving out any mirror flags:

git clone https://github.com/myshell-ai/MeloTTS.git
cd MeloTTS
pip install -e .
python -m unidic download

Use a virtual environment or a Conda environment for this. The requirements file pins transformers to 4.27.4 and includes torch, torchaudio, gradio, librosa 0.9.1, pydub, and a stack of language-specific text processors: unidic, unidic_lite, mecab-python3, fugashi, pykakasi, g2p_en, pypinyin, jieba, and g2pkk. That is a lot of pinned dependencies to install into a shared Python. A dedicated environment is the difference between a clean rollback and a broken system Python.

The fourth command is required, not redundant. The package's setup.py defines a post-install hook for the dictionary download but never registers it, and the project's own Dockerfile runs python -m unidic download explicitly. unidic is the dictionary the Japanese text path uses.

Once installed, the web interface starts with melo-ui or python melo/app.py. The command line tool is invoked as melo or melotts. The documented examples are short:

melo "Text to read" output.wav
melo "Text to read" output.wav --language EN --speaker EN-US
melo "Text to read" output.wav --language EN --speaker EN-AU
melo "Text to read" output.wav --speed 1.5
melo file.txt out.wav --file

For scripted work, the Python API is the more useful surface. The documented pattern creates a TTS object for a language, reads the speaker map, and writes a wav file:

from melo.api import TTS

model = TTS(language='EN', device='cpu')
speaker_ids = model.hps.data.spk2id
model.tts_to_file("Did you ever hear a folk tale about a giant turtle?", speaker_ids['EN-US'], 'en-us.wav', speed=1.0)

The install docs' example uses device='auto', with a comment that it will use the GPU if one is available. On a Mac, force device='cpu'. The reason is in the next section.

Where the Mac path breaks

The install docs' comment lists mps as a valid device value alongside cpu and cuda. Users have not found that reliable. Issue #228, opened in January 2025 on an M3 Pro running macOS 14.7, reports this failure during synthesis: "Output channels > 65536 not supported at the MPS device." A second commenter reproduces it on an M1 MacBook Pro with torch 2.5.1 and links a PyTorch issue and pull request about the same operator limitation.

The workarounds in that thread are the ones you would expect. Set PYTORCH_ENABLE_MPS_FALLBACK=1 so unsupported operations drop back to CPU, or pass device='cpu' and skip MPS entirely. Neither gives you Metal acceleration for the failing operation. For a Mac user, the practical reading is that the GPU path is not a supported feature of this project today.

There is an older macOS failure too. Issue #76, opened in March 2024, ends in an IndexError inside the attention code: "Dimension out of range (expected to be in range of [-3, 2], but got 3)." Comments report the same traceback on an M1 MacBook, an M2 mini, and an M3 Mac Pro, through both the CLI and the web UI. That issue is still open. It may be fixed by a dependency combination that works on your machine, and it may not. This is the kind of uncertainty the README's Docker suggestion is trying to route around.

Docker is a reasonable fallback on macOS, with one trap. The project's run command in the GPU variant is docker run --gpus all -it -p 8888:8888 melotts. That flag asks for NVIDIA hardware and will fail on a Mac. Use the plain docker run -it -p 8888:8888 melotts form, then open localhost:8888. The install docs also say the image build can take a few minutes.

Two more practical details. The requirements include cached_path, and the Dockerfile runs python melo/init_downloads.py, which is how pretrained checkpoints get fetched. Expect a download the first time you synthesize, and plan disk space for one checkpoint per language you use. Separately, if pip has to build the MeCab and fugashi dependencies from source, you may need a compiler toolchain on the machine.

License, consent, and what MIT does not settle

The repository LICENSE is the MIT License, copyright 2024 MyShell.ai, and the README says it is free for both commercial and non-commercial use. The Hugging Face cards for the English checkpoints also list mit. That is about as permissive as a model license gets, and it is a genuine advantage over projects with research-only or non-commercial terms. Read the license file yourself if you need to make a compliance decision, and keep the copyright notice with any substantial portion you redistribute.

MIT covers the code, and the Hugging Face cards for the English checkpoints also declare MIT. It says nothing about the provenance of the training audio or the rights of the people who recorded it. It also does not grant you rights to a voice you imitate. If you fine-tune on a client's recordings, get written permission and keep it with the project. The open source model license checklist covers the questions worth asking before you ship output commercially.

On cloning, be precise about what the project offers. The README documents preset speakers and a training path, not zero-shot cloning from a reference clip. There is no argument in the documented API for a reference wav and its transcript, which is the signature of a cloning model. The training document does show how to fine-tune on a custom dataset: a metadata file with audio path, speaker name, language code, and transcript, then preprocess_text.py, then train.sh with a GPU count, then infer.py against the resulting checkpoint. That is a real customization route, but it needs transcribed audio, training time, and hardware, and it is not the same thing as dropping in a ten-second sample.

Only clone voices you own or have written permission to use. That rule applies to fine-tuning as much as to any commercial cloning product. A local model removes the upload step, not the consent requirement.

Should you run it yourself?

Your situation	Self-hosted MeloTTS	A packaged Mac app
You specifically need MeloTTS English accents, including Indian or Australian	Good fit. Those speaker IDs are in the English checkpoint.	Check the app's voice list first. Kokoro in Murmur covers US and UK English, not the full MeloTTS accent set.
You want a short, light, CPU-friendly voice for drafts	Works. The install docs say CPU is sufficient for real-time inference.	Kokoro is bundled in Murmur and runs locally on Apple Silicon and Intel Macs with preset voices.
You need zero-shot cloning from a reference clip	Not the documented feature. Fine-tuning is the only customization path in the docs.	Murmur ships several reference-cloning models. Confirm the current list and requirements in the app.
You do not want Python, Conda, or a terminal	Poor fit. Install and troubleshooting are yours.	Fits. Native macOS app for Apple Silicon and Intel Macs, macOS 15 or newer.
You need a Linux server or batch job	Fits. The project is tested on Ubuntu and exposes a CLI and Python API.	Not a fit. It is a desktop Mac app.
You need Windows	The project offers a Docker path, but it is outside this Mac guide.	Not available. macOS only.
You need live streaming into a call or game	Not a documented streaming product. You would be building around a file API.	Not a live streaming product either. It generates and edits files.

The honest summary: run MeloTTS yourself if its accent set or mixed Chinese and English behavior is the reason you are here, or if you want to study and modify the model. Do not pick it if you want a finished editor, a supported Metal path, or voice cloning from a short sample. The studio versus DIY comparison goes through that trade in more detail.

What can go wrong

The install is tested on Ubuntu 20.04 with Python 3.9. On macOS you are on an untested path, and the project's own fallback advice is Docker.
device='auto' can select MPS and fail with the output channels error from issue #228. Pass device='cpu' or set PYTORCH_ENABLE_MPS_FALLBACK=1.
Some macOS installs fail inside the attention code with a dimension IndexError, reported in issue #76 on M1, M2, and M3 machines and still open.
The GPU Docker command uses --gpus all, which is an NVIDIA flag. On a Mac, use the plain docker run form.
Pinned dependencies collide with other projects. transformers is held at 4.27.4, and the package needs its own environment.
Native text processing dependencies may need a compiler if no wheel matches your Python version.
Checkpoints download at first use, and you need one per language. Budget disk space before a multi-language project.
The project documents preset speakers and fine-tuning, not zero-shot cloning. Do not plan a cloning workflow around it.

The packaged alternative, and when it is the wrong call

If the appeal of MeloTTS is small, local, CPU-friendly speech rather than its specific accents, a packaged app removes the Python stack entirely. Murmur is a local voice studio for Apple Silicon and Intel Macs running macOS 15 or later. Kokoro is bundled, so the first voice works without a model download, and the app handles scripts, multiple speakers, takes, and export in one place. Kokoro covers US and UK English plus Spanish, French, Hindi, Italian, Japanese, Portuguese, and Chinese, with preset voices rather than cloning.

Murmur also ships larger models for other jobs: Chatterbox Turbo, Chatterbox Multilingual, Qwen3-TTS Base, Voice Design, SparkTTS, Fish Audio S2 Pro, Breeze TTS 2, and OmniVoice. Those cover reference cloning and multilingual work that MeloTTS does not document. What Murmur does not ship is MeloTTS itself, and no setting in the app will make it appear. If a client asks for the exact EN_INDIA voice from that checkpoint, the app cannot give it to you.

That is the fair boundary. MeloTTS wins on one narrow thing: a permissive license and a fixed set of multilingual speakers that run on CPU. A packaged app wins on everything around the model, which is usually where projects stall. If you want the no-terminal route, the no Python local TTS guide walks through it, and the Kokoro on Mac guide covers the bundled model in depth. If you are still comparing open models, start with the local TTS model guide.

Frequently asked questions

Does Murmur run MeloTTS?

No. Murmur ships Kokoro, Chatterbox Turbo, Chatterbox Multilingual, Qwen3-TTS Base, Voice Design, SparkTTS, Fish Audio S2 Pro, Breeze TTS 2, and OmniVoice. MeloTTS is not in that catalog. To use MeloTTS you install the official Python project yourself.

Can MeloTTS run on an Apple Silicon Mac?

Yes, with caveats. The Python package can be installed and run on CPU on Apple Silicon. The project tests on Ubuntu 20.04 with Python 3.9, does not document macOS as supported, and its own install page points macOS users with problems to Docker. The Metal path is not reliable: issue #228 reports an MPS operator error that forces a CPU fallback.

Does MeloTTS support voice cloning?

The README documents preset speakers and a fine-tuning path, not zero-shot cloning from a reference clip. Fine-tuning needs transcribed audio, preprocessing, and training time, and the training doc is written around GPUs. If you want cloning from a short sample, MeloTTS is the wrong tool.

Which languages and accents does MeloTTS include?

English with five accents (default, US, British, Indian, Australian), Spanish, French, Chinese, Japanese, and Korean. The Chinese speaker supports mixed Chinese and English text. Each non-English language has a single speaker in the documented examples.

Can I use MeloTTS output commercially?

The repository and the Hugging Face English cards state the MIT License, and the README says it is free for commercial and non-commercial use. MIT does not settle the rights in your training or reference audio, so confirm you have permission for any voice you imitate. This is not legal advice.

Sources

Try Murmur - $49 one-time. No subscriptions, no cloud, no per-character fees.

Originally published at murmurtts.com

Kindle Text to Speech: What Reads Aloud and What Does Not

Tarun yadav — Mon, 09 Nov 2026 17:55:19 +0000

Which Kindle devices and apps read books aloud, which books Amazon excludes, and how to make audio from DRM-free EPUB or PDF files on a Mac.

Direct answer: Kindle text to speech is built into three places. Assistive Reader runs on newer Kindle e-readers and inside the free Kindle apps, VoiceView runs on Fire tablets and Kindle devices as a full screen reader, and Apple's Speak Screen and Spoken Content can read whatever is on your iPhone, iPad, or Mac screen, including a Kindle page. All of them are free and read eligible books aloud, but none of them hands you an audio file, and none of them will read every title.

That last point is where most searches go wrong. People ask whether Kindle has text to speech, get a yes, then discover their specific book will not play, or that they cannot take the narration anywhere. This guide separates the routes Amazon and Apple actually document, shows which books are excluded, and then covers the legal alternative: turning a file you own into a finished audio file on a Mac.

The read-aloud routes Amazon and Apple actually document

Assistive Reader on a Kindle e-reader. Amazon says Assistive Reader provides text-to-speech controls on Kindle E-Reader 11th Generation and higher, Kindle Scribe, and Kindle Colorsoft. It supports books in English, Spanish, French, German, and Italian. You get play, pause, reading speed, and a 30-second rewind, plus word highlighting as the book is read. It pauses by itself when you open the navigation menu or make a highlight or note. It only works with books that have Enhanced Typesetting, a setting you can see in the Kindle Store feature list.

Assistive Reader inside the Kindle apps. The free Kindle app for iPhone, iPad, Android phones and tablets, Mac, and Fire tablets can read books aloud too, using the device's operating system text-to-speech engine rather than a voice Amazon supplies. Amazon lists the eligible sources: books purchased from the Kindle Store, borrowed from Kindle Unlimited, Prime Reading, or other official Kindle partner services, or shared through a Family Library. The book also has to be in a language your device can speak. Amazon's numbered setup steps name iOS, Android, and Fire tablet; the Mac and iPad are listed as supported apps but are not in the step list.

VoiceView. This is a full screen reader, not just a play button. On Fire tablets it works on FireOS 5 and later with continuous or item-by-item reading, and you can change the voice under Settings, Keyboard and Language, Text-to-Speech. VoiceView reads in your device language, and Amazon notes that not every language is supported.

Apple Speak Screen and Spoken Content. These belong to the operating system, not to Kindle. Apple documents settings under Accessibility, Read and Speak, including an Accessibility Reader with Autoplay, a two-finger swipe down from the top of the screen for Speak Screen, and Speak Selection for highlighted text. On a Mac, Speak selection has a default shortcut of Option-Esc plus highlight and controller options. Because it reads the screen, it can read a Kindle book page even where Amazon's in-app feature is missing, but the voice and the controls are Apple's.

Decision aid: match the route to what you need

Route	Where it runs	What it reads	What you get
Assistive Reader (e-reader)	Kindle E-Reader 11th Gen and later, Kindle Scribe, Kindle Colorsoft	Kindle books with Enhanced Typesetting, in five languages	In-app playback with speed, 30-second rewind, word highlighting
Assistive Reader (app)	Free Kindle app on iPhone, iPad, Android, Mac, Fire tablets	Purchased, borrowed, or Family Library books with Enhanced Typesetting, in a language your device speaks	In-app controls and highlighting, voice from the device OS
VoiceView	Fire tablets on FireOS 5 and later; check Amazon's Help for your Kindle e-reader model	Kindle books, read continuously or item by item	Screen reader navigation plus text to speech
Apple Speak Screen	iPhone, iPad, Mac	Text on screen, including a Kindle book page	OS playback with rate, highlighting, and a controller
Your own audio file	Apple Silicon or Intel Mac with macOS 15 or later	DRM-free EPUB and PDF you own or wrote	A WAV or M4A file with chapter markers

Which Kindle books will not be read aloud

Enhanced Typesetting is the gate. Amazon says Assistive Reader works with any Kindle book that has Enhanced Typesetting, and the way to check is the book's page in the Kindle Store, where the features list shows either "Enhanced Typesetting: Enabled" or nothing. Amazon's accessibility page adds the harder limit: print replica and image-based fixed layout Kindle books are not readable in read aloud and braille. That category includes print replica, manga, comics, and some children's books and textbooks. The same page notes two more gaps, that alt text and MathML are not currently readable on Kindle e-readers, and that Kindle books do not support closed captions in audio or video content.

A book without Enhanced Typesetting, which you can check in the Kindle Store feature list.
Print replica, comics, manga, or another image-based fixed layout title.
A book in a language your device's text-to-speech engine does not speak.
A title you did not purchase, borrow through an official Kindle source, or receive through a Family Library.
Anything you only have as a purchased Kindle file, which is locked to Amazon's formats and apps.

Turning on Assistive Reader, step by step

On a supported Kindle e-reader, per Amazon:

Open the book you want to hear.
Tap the top center of the screen.
Select the reading settings menu, shown as Aa.
Select More, then turn on Assistive Reader.
Tap the screen to show the player controls, then set the reading speed or rewind 30 seconds.

In a Kindle app on iOS, Android, or a Fire tablet, the flow is nearly the same, with one extra shortcut. Open the app and the book, tap the top center, then Aa, then More, then turn on Assistive Reader. Tap the screen to show the player controls. To start from a specific sentence instead of the top of the page, long press a word to open the quick actions menu and select Play. Once Assistive Reader is on, Amazon says it stays available until you turn it off again.

VoiceView and Speak Screen: when to use the accessibility route

Use VoiceView when you want to navigate the whole book, not just listen to the text. On a Fire tablet, Amazon's Help documents continuous reading of every page or non-continuous reading of one page, character, or word at a time, with a reading toolbar, heading navigation, and a way to change the voice.

Reach for Apple's tools when the in-app feature is not available on your device. Turn on Speak Screen in Settings under Accessibility, Read and Speak, then swipe down with two fingers from the top of the screen, or select text and use Speak Selection. The same settings control the speaking rate, whether content is highlighted as it is read, and the on-screen controller. This is the practical fallback for reading a Kindle page on a device where Assistive Reader does not appear.

What these voices sound like, according to the docs

Amazon describes Assistive Reader as using your device's operating system text-to-speech engines, which means the voice comes from the hardware, not from a narrator recording. The pages do not promise a specific voice, accent, or quality level. On Kindle e-readers, Assistive Reader supports five languages, and VoiceView reads in the device language with the caveat that not all languages are supported. Apple's Speak Screen uses the system voices installed on your device, and you can change the rate, highlighting, and controller behavior. The honest takeaway is that none of these routes is a performance; they are utility voices you should audition on your own hardware before you commit to a long book.

Where built-in read-aloud stops being enough

The moment your goal shifts from "let me listen" to "I need audio," the built-in routes run out of room. They do not produce a file. They do not let you fix a mispronounced name, choose a second voice for a character, or save a chapter as M4A for a commute, a class, or a client. They keep the playback inside one app on one device, so a listener without that app gets nothing. And for a purchased book, they give you listening rights inside Amazon's ecosystem, not a distributable master.

There is a rights question too, separate from the technical one. Listening to a book you own is one thing. Making a permanent audio version and giving it to other people, publishing it, or selling it is a licensing decision that belongs to the copyright holder. Check the rights before you export anything meant for an audience.

The legal alternative: make audio from files you can actually open

If you own a DRM-free EPUB or PDF, or you wrote the manuscript yourself, you can generate narration on a Mac. Murmur is a local-first voice studio for Apple Silicon and Intel Macs running macOS 15 or later. It imports a PDF or an EPUB, reads the text in document order, splits it into chunks, and exports WAV or M4A. When a chaptered timeline is exported as M4A, the chapter markers travel with the file. Two things to be clear about: Murmur cannot open Kindle AZW or KFX files, and this guide does not explain how to remove DRM. If your only copy is a purchased Kindle book, use the built-in read-aloud routes above instead.

The EPUB to audiobook workflow covers spine order, chapter scripts, and why regeneration matters. The e-reader text to speech comparison covers when to stay with the reader's own voice. For delivery, see converting an ebook to M4B and the M4B and MP3 format guide. If all you need is for the Mac to read a page to you, the Speak Selection guide is the lighter tool.

A worked example: one chapter from a DRM-free EPUB

Picture a twelve-chapter novel you own as an unprotected EPUB. You want a narration you can play in the car, and you want to fix pronunciation before committing hours to a full run. Here is the order that saves the most time, and it works the same for a PDF with a clean text layer.

Confirm the file opens without a password and is genuinely EPUB or PDF. Murmur accepts those two formats and will not open a Kindle AZW or KFX file.
Import the EPUB and read the extracted preview. Murmur follows the book's spine, so the chapters should arrive in reading order.
Trim the front matter. Copyright pages, indexes, and stray page numbers do not need to be spoken.
Check names, places, and invented terms before you generate. One wrong pronunciation repeats in every chapter if you fix it late.
Generate chapter one only, listen end to end, and adjust the script or the voice before you continue.
Generate the remaining chapters with the same voice and settings, then export M4A for chapter markers or WAV if you need a lossless master.

For the voice itself, Kokoro is bundled with Murmur and runs on any supported Mac. The rest of the model catalog, which includes Chatterbox, Qwen3-TTS, Voice Design, and several others, needs a Mac with 16GB of unified memory or more. Check the model list in the app for the current set.

When Murmur is not the right choice

Murmur runs only on Macs (Apple Silicon or Intel) with macOS 15 or later. It is not a Windows, iPhone, Android, or browser app, it does not do live or real-time streaming speech, and it does not do speech to text. It is not a cloud service, so there is no shared workspace for several editors. If you need any of those, use a tool that has them. And if your only goal is to hear a Kindle purchase while you commute, Assistive Reader or Speak Screen is already the simplest answer.

Frequently asked questions

Can Kindle read any book aloud?

No. Assistive Reader needs a book with Enhanced Typesetting, which you can check in the Kindle Store feature list. Amazon also says print replica and image-based fixed layout books, including comics, manga, and some children's books and textbooks, are not readable in read aloud and braille, and the book has to be in a language the device's text-to-speech engine supports.

Does Kindle text to speech work on a Mac?

Amazon lists the free Kindle app for Mac among the apps that support Assistive Reader, but its numbered setup steps name only iOS, Android, and Fire tablet. On a Mac you can also use the operating system's Speak selection, which Apple documents under Accessibility, to read a page on screen. If neither appears, generate your own audio from a DRM-free EPUB or PDF.

Can I save the Kindle read-aloud narration as an audio file?

No. Assistive Reader, VoiceView, and Apple Speak Screen all play audio through the device. None of them exports a WAV or M4A file you can copy, share, or distribute. Turning text into a saved audio file is a separate workflow.

Can Murmur open a Kindle AZW or KFX file?

No. Murmur imports PDF and EPUB only. It cannot read Kindle formats, and this guide does not cover removing DRM. Use Assistive Reader for a purchased Kindle book, or generate audio from a DRM-free EPUB or PDF you own or wrote.

What is the difference between Assistive Reader and VoiceView?

Assistive Reader is a reading feature with a play button, speed control, 30-second rewind, and word highlighting. VoiceView is a full screen reader that also lets you navigate the interface and move through a book item by item. VoiceView is the better fit when you need to operate the device, not just listen.

Do I need to remove DRM to make an audiobook from a book I own?

No. Use a DRM-free EPUB or PDF, a public domain text, or a manuscript you wrote. If your only copy is a purchased Kindle book, stick with the built-in read-aloud features, because making and distributing an audio version is a rights decision you should confirm with the copyright holder.

Sources

Try Murmur - $49 one-time. No subscriptions, no cloud, no per-character fees.

Originally published at murmurtts.com

Screen Recording Voiceover: Script and Narrate on Mac

Tarun yadav — Sun, 08 Nov 2026 14:38:31 +0000

Add a screen recording voiceover on a Mac: capture the screen silently, script it scene by scene, then lay a separate narration track under the video.

Direct answer: to add a voiceover to a screen recording on a Mac, record the screen silently first, write a script that follows what happens on screen, then produce the narration as a separate audio track and place it under the video. Keeping the picture and the voice as two files means you can fix one line, change the voice, or retime a scene without capturing the screen all over again.

Most how-to pages treat a screen recording voiceover as something you talk over the capture. That works for a quick clip, but it ties the words to the take: a stumble means a new take, and a changed fact means re-shooting the whole demo. The alternative is to split the job in two. Capture the screen as a silent movie, write the script against that movie, then generate or record the narration as its own track and lay it under the picture. This is also how a product demo script is usually produced, and it is the workflow this guide walks through end to end.

Record the screen first and narrate second

Choose the method before you start, because the cleanup is different for each one.

Approach	What you do	Strengths	Weaknesses
Narrate live while recording	Talk through the steps as the screen records, capturing voice and picture in one pass.	Fastest for a personal one-take clip, and the timing is automatic because you speak as you click.	A stumble means redoing the take, the mic also picks up room noise, and you cannot fix one sentence without re-shooting.
Record silent, add a narration track	Capture the screen with the microphone off, then write and produce the voice as a separate audio file.	The script can be written, reviewed, and regenerated; one line can change without re-recording the screen.	You have to plan the timing and sync the track yourself.

The separate-track method is the one that scales to a tutorial, a course module, or a support walkthrough, because the two parts change at different rates. Screen actions are stable; wording is not. When a control label changes, you edit one line of the script and regenerate one audio file instead of re-shooting the demo. A live take is still the right choice for a casual update where the point is your voice reacting in the moment.

Script the demo from the screen actions

Write the narration against the capture, not against your memory of the app. Watch the silent recording once and note where the cursor lands and what changes on screen, then fill one short block per scene. Keep the words the audience must hear in the spoken script, and keep production directions such as zoom in here out of it.

Script beat	What the narration says	What the capture shows
Goal	One sentence naming the task the viewer has.	The starting screen, before anything changes.
Setup	One sentence naming the file, project, or account in use.	The open window or selected item.
Action	The exact control name and what happens when it is used.	The click or selection, without a cut in the middle.
Result	What the viewer should now see.	The changed state, held long enough to read.
Boundary	What this demo does not cover.	The context needed to make the limit clear.
Next step	One action for the viewer to take.	The destination or final instruction.

A useful test: cover the video and read only the spoken column. If the words still work as a set of steps, the narration can stand on its own. If it only makes sense while pointing at the screen, replace here and this with the control name and a description of the result. Name the panel, the button, and the state that appears, because a viewer following along on their own Mac cannot see your cursor.

Here is a short three-scene script written against a silent capture. Each scene maps to one continuous take.

Scene 1, New project
In this clip we create a project and add one audio file. Start on the home screen, where saved projects appear as cards. Choose New project, name it Sample greeting, then choose Create.

Scene 2, Add the narration
Open Add audio, then pick the file sample-greeting.wav from the folder on the left. The clip drops onto the timeline at the playhead.

Scene 3, Export
With the clip in place, choose Export. Pick the format and the location, then start the export. A progress bar runs across the top until it finishes.

This demo shows adding one file and exporting it. It does not cover trimming, multi-track mixing, or any cloud feature.

Record a silent screen capture on a Mac

macOS records the screen with the Screenshot tool, and you can open it from the keyboard or from QuickTime Player. For a narration workflow, record with the microphone set to None so the file holds picture only, then add the voice later.

Press Shift-Command-5 to open the Screenshot toolbar. In QuickTime Player you can also choose File > New Screen Recording to open the same tools.
Pick a capture target: record the entire screen, a window, or a selected portion. Drag the frame to reposition it, or drag its edges to resize the area.
Open the Options menu and set Microphone to None so no microphone audio is captured. This is the step that keeps the recording silent.
In the same Options menu, choose where the file saves, set a timer delay if you need a moment before recording begins, and turn on Show Mouse Pointer if the clicks matter to the viewer.
Click Record, perform the workflow at a steady pace, then click the Stop Recording button in the menu bar. Screen recordings save as .mov files.

Record the whole workflow once, even if you expect to cut it. A silent capture costs nothing to re-trim, and the extra footage gives you room to match a line of narration to the exact frame where a change appears. Keep cursor movement deliberate: a pointer that wanders makes the viewer guess which control you mean. If the demo has a step that is hard to reproduce on demand, a quick rehearsal before the real take saves more time than a re-record.

Produce the narration track scene by scene

Once the screen capture exists, the script becomes audio. Generate or record one file per scene and name each file after the scene, so a later revision maps to the right place in the timeline. Murmur is a local-first macOS voice studio for Apple Silicon and Intel Macs running macOS 15 or later. It handles text to speech, voice cloning, voice design, and multi-speaker projects, and it runs the models on your own machine, so there are no per-character fees and nothing is uploaded to a cloud service. That matters when the demo shows an unreleased screen or a client name. Murmur costs $49 one-time and has a free plan and a 3-day Pro trial.

Split the approved script into one block per scene and keep the file next to the project.
Generate the audio, then export a format your editor accepts, such as WAV or M4A. The audio export guide explains master versus delivery files.
Name each file to match its scene, for example 01-new-project.wav, so the files sort in order in the Finder.
Listen to each clip against its scene and note any line that lands late or early, then fix the script rather than the timeline where you can.
Keep the clean audio files outside the editor so you can swap one scene without disturbing your edits.

Use one voice for the whole demo, chosen before you generate anything. A voice that changes between scenes reads as a mistake even when the words are identical. If you are still choosing, the narration voice roundup compares the options, and the Mac narration workflow covers the script to audio handoff in more depth. Export a consistent level and sample rate across scenes so the joins do not jump in volume, and keep the master files outside the editor.

Sync the narration under the video

Place the narration under the video, not in a background music well, so it moves with the picture when you trim a scene. In iMovie, drag the audio file right under the video clip it belongs to, and a connecting bar shows the two are attached. A file dropped in the music well stays independent and will not follow a video edit.

Drop each scene's audio under the matching video clip so the two stay attached.
If a clip runs long, trim the silence at its head or tail before you move it, so the first spoken word lands where the action starts.
Select the narration clip and lower the volume of other clips so any music or system sounds duck under the voice.
Watch the whole demo on the speakers your audience will use, then check every join, because an audio clip can start a beat early or end with a tail of silence.

If the recording needs to go to a platform video rather than a standalone tutorial, the AI voiceover workflow covers publishing the finished piece, and the audio export guide explains the difference between an editing master and a delivery file. For a longer training build, the training video narration guide covers the wider production loop.

When Murmur is not the right choice

Murmur is a Mac app, so if you edit on Windows, need live or real-time streaming speech, or want several people editing one cloud project at once, a cloud service will suit you better. Tools that combine recording, hosting, and transcription in a browser, such as Loom or Descript, are a better single-tool fit for that workflow, so compare their plans directly. Murmur also does not do speech to text, so it cannot turn a recorded walkthrough into a script. And if the demo needs a specific celebrity voice or an emotional read that a stock generated voice cannot carry, plan on recording a person instead. Murmur is strongest when you want private, repeatable narration you can regenerate from the same script, scene after scene.

Frequently asked questions

How do I add a voiceover to a screen recording on Mac?

Record the screen silently first, then add the narration as a separate audio track. On a Mac, press Shift-Command-5 to open the Screenshot tools, set Microphone to None so the recording holds no voice, capture the workflow, and then place a generated or recorded narration file under the video in your editor.

Should I talk while recording the screen or record the voice separately?

Talking live is fastest for a one-take clip, because the timing is automatic. A separate track is better when the narration has to be repeatable, reviewed, or revised, because you can change one line and regenerate one file without recapturing the screen. Pick the method based on how likely the wording is to change.

Where does macOS save a screen recording?

Screen recordings save as .mov files, and the Screenshot Options menu sets the save location. The filename begins with Screen Recording and includes the date and time. You can change the folder, or choose a preset location, before you click Record.

Why is my narration out of sync with the screen?

A long take that was never split, or an audio file dropped in a background music well instead of under a video clip, is the usual cause. Attach each scene's audio under its clip, trim the silence at the head of the clip, and nudge it until the first spoken word lands where the action starts.

Can Murmur record the screen for me?

No. Murmur produces the narration audio; the screen capture comes from macOS or another screen recorder. Record the screen with the Screenshot tools or QuickTime Player, then use Murmur to make the voice track that you lay under the video.

Sources

Try Murmur - $49 one-time. No subscriptions, no cloud, no per-character fees.

Originally published at murmurtts.com

Mac say Command: Flags, File Output, and Quality Limits

Tarun yadav — Sat, 07 Nov 2026 14:29:20 +0000

A tested guide to the macOS say command: voice and rate flags, writing AIFF, CAF and M4A, the WAV failure, and the 22 kHz quality limit to plan around.

Short answer: The macOS say command converts text to speech straight from Terminal. Run say -v Samantha -o notes.aiff -f script.txt to read a file and write an audio file, or say "Ready" to just play the words aloud. The -v flag picks the voice, -r sets words per minute, and -o writes AIFF by default, and the tool ships with every Mac, so it needs no install or account.

Everything below was tested on macOS 27 (build 26A434) with the built-in /usr/bin/say. The default output measured mono, 22050 Hz, 16-bit, and a plain .wav path failed until an explicit data format was supplied. Voice lists and file behavior change between macOS releases, so confirm the details on your own machine before you build a workflow on them.

What the say command actually does

say is a command-line front end to the same on-device speech synthesis that Mac apps reach through Apple's AVFAudio framework. Apple documents that speech synthesis creates an utterance, configures a voice and rate, and passes it to a synthesizer that produces audio. The say tool exposes the common parts of that pipeline without any code. It reads input from three places: a string on the command line, a file given with -f, or standard input when you pipe text in. It then either plays the result through your chosen output device or writes it to a file.

Task	Command	What happens
Speak a string	say "Build finished"	Plays through the default output device
Speak a file	say -f script.txt	Reads the file and speaks its contents
Read from a pipe	cat script.txt \| say	Speaks text received on standard input
Write audio	say -o take1.aiff "Take one"	Writes an AIFF file instead of playing
List voices	say -v '?'	Prints every installed voice and language
List formats	say --file-format='?'	Prints the writable file formats

Notice that playing and writing are the same action with a different destination. If you omit -o, the audio goes to your speakers and disappears. If you provide -o, you get a file and nothing is heard. That single distinction is what makes say useful both as a notification tool and as a rough audio generator.

The flags that matter most

The default voice is whatever you selected in System Settings, so a script that names a voice explicitly is more predictable for automation. Rate is set in words per minute with -r. In a test on the same sentence, 120 wpm produced about 3.69 seconds of audio while 240 wpm produced about 2.58 seconds, so rate changes are easy to hear and easy to measure. The same sentence always produces the same length at a fixed rate, which makes say handy for estimating how long a draft will run.

-v voice selects a voice. Use -v '?' to print the installed list, which varies by macOS version.
-r rate sets the speech rate in words per minute.
-o file writes audio to a file instead of playing it.
-f file reads text from a file. Pass -, or pipe text in, to read from standard input.
-a device routes playback to a named audio device.
-n name:port sends speech over the network through AUNetSend.
--data-format and --file-format control the container and codec.
--progress prints a progress meter during synthesis.
-i highlights and speaks text line by line as it is synthesized.

say -v '?' | head
say -v Samantha "Listen to this sentence"
say -v Samantha -r 150 -o take.aiff -f script.txt
say -v Samantha -o take.m4a --data-format=aac -f script.txt

You can also insert pauses inside the text. Apple's own examples use inline commands such as [[slnc 1500]] to hold for 1.5 seconds. In a test, adding that command to a two-word phrase grew the clip by roughly 1.3 seconds, so the markup works, but it is Apple-specific and not a portable standard. If you need pauses to survive a move to another engine, keep a plain-text master and add pauses as a separate step.

Output to a file, and the WAV trap

The safest output is AIFF. The file extension usually selects the container, but not every combination is accepted. On the tested build, AIFF, AIFC, CAF, and M4A all wrote without extra flags, while a plain .wav path failed with Opening output file failed: fmt? until a data format was added. AIFF also rejected the raw @samplerate syntax that CAF accepted, which is a useful warning against assuming the flags behave the same on every container.

Output file	Result on the tested build	Notes
voice.aiff	Works. AIFF, mono, 22050 Hz, 16-bit	Big-endian PCM and the safe default
voice.caf	Works. Core Audio Format	Accepted --data-format=LEI16@44100 to request 44.1 kHz
voice.m4a	Works. AAC, mono, 22050 Hz	Add --data-format=aac for compressed audio
voice.wav	Failed with a fmt? error	Add --data-format=LEI16@22050 to write a WAV file

say -v Samantha -o take.aiff -f script.txt
say -v Samantha -o take.caf --data-format=LEI16@44100 -f script.txt
say -v Samantha -o take.m4a --data-format=aac -f script.txt
say -v Samantha -o take.wav --data-format=LEI16@22050 -f script.txt

Always confirm what you actually wrote instead of trusting the request. Run afinfo take.aiff, or ffprobe take.m4a if FFmpeg is installed, and read the channel count, sample rate, and codec from the header. A file that exists is not proof that it is the format you asked for. This check takes a second and prevents a wrong-format file from reaching an editor or a phone system.

Quality limits to understand before you ship

It helps to know what say is not. The Apple voices are convenient and free, but the output is a system text-to-speech engine, not a narration studio. The following limits come from the official man page and from measurement, and they define where say stops being enough.

Output is mono by default at a 22050 Hz sample rate on the tested build.
SS64's say reference notes that many macOS voices are sampled at around 11 kHz, so resampling higher does not recreate detail that was never captured.
Novelty voices such as Bells, Boing, and Bubbles ship with the system but are not narration voices.
There is no SSML. Apple's inline commands, such as [[slnc]] for a pause, are the available markup and they are specific to this engine.
Expression control is limited to voice and rate. You cannot request a mood, a style, or an emotion.
The --bit-rate and --quality options frequently just re-encode or upsample. SS64 warns that they can increase file size without improving the sound.

I can measure a say file, but I cannot tell you whether the voice suits your project. Generate a short sample, listen on the speakers or headphones your audience will use, and decide from the audio. Do not ship a long narration you never auditioned, and do not assume a higher requested sample rate fixed a thin or robotic source.

When say is the right tool, and when it is not

Use say when the job is quick, scriptable, and does not need a performance. Reach for a dedicated engine or a local model when the audio itself has to carry the message.

Choose say when	Choose something else when
You want spoken alerts from a shell script, cron job, or CI step	You need natural, expressive narration for a published video or audiobook
You need read-aloud on a locked-down Mac with no installs allowed	You need voice cloning or a specific brand voice
You want a rough timing and pronunciation preview before committing	You need multi-speaker projects assembled and exported as one file
You want zero cost, zero account, and offline operation	You need Windows, Linux, iPhone, Android, or a cloud pipeline
The text is short and subtle prosody does not matter much	You need fine control over pacing, emphasis, or names at scale

The strength of say is that it is already there. There is nothing to license, download, or configure, and it runs entirely on the machine. That is exactly why it is a bad fit when you need emotion, cloning, or a repeatable house voice, because those are qualities the system engine was never designed to deliver.

A practical script-to-audio workflow

Write the script in a plain text file, one paragraph per thought, so you can regenerate a single section later.
List the installed voices with say -v '?', then record a short sample with two or three candidates at your intended rate.
Set the rate deliberately. A slower rate is easier to follow for accessibility, while a faster rate fits more words into a fixed window.
Generate one file per section. If a word is mispronounced, rewrite that word or spell it out and regenerate only that section.
Join the sections with your editor or a tool such as FFmpeg, then export the delivery format your destination expects.
Keep the script next to the audio so a later edit maps back to the exact line that changed.

The Mac text to speech workflow covers the script-to-audio handoff in more depth, and the WAV, M4A, and MP3 export guide explains the difference between a master file and a delivery copy. If the job grows past a few files, the local TTS automation and batch guide walks through job queues, idempotency, and safe retries.

Where a local voice studio goes further

When a rough system voice is not enough, a local model app is the next step. Murmur is a local voice studio for Apple Silicon and Intel Macs running macOS 15 or later. It bundles Kokoro, a fast preset-voice model with 10 or more built-in voices, and it can also run larger cloning and voice-design models, which want 16 GB of memory or more. Generation happens on your Mac with no per-character fees and no cloud upload. Murmur costs $49 one-time, has a free plan and a 3-day Pro trial, and offers a 7-day refund policy, so audition the published samples before you buy.

It is not the right tool for every situation. If you need Windows, an iPhone or Android app, a browser interface, or live streaming speech, Murmur does not provide those, and it does not do speech-to-text. For a quick notification sound or a throwaway draft, say remains the simpler answer, and the two tools are not really competitors for the same job.

For a wider comparison of reading and production apps, see best text to speech apps for Mac. For a no-code route to daily read-aloud, the Mac text to speech shortcut guide builds the same capability into Shortcuts.

Frequently asked questions

What is the macOS say command?

say is the built-in Terminal tool that converts text to speech using Apple's system voices. Give it a string, a file with -f, or piped text, and it plays the words or writes an audio file with -o. It requires no install and runs on the Mac itself.

How do I save say output to a file?

Add -o followed by a path, for example say -o take.aiff -f script.txt. AIFF is the default and the safest choice. For compressed audio use -o take.m4a --data-format=aac. Confirm the result with afinfo rather than assuming the format.

Why did my say .wav file fail to write?

On the tested build, a plain .wav path failed with a fmt? error while other extensions worked. Supplying a data format, for example --data-format=LEI16@22050, produced a valid WAV file. The accepted combinations differ by container and macOS version, so test on your machine.

Can say use SSML or sound more expressive?

There is no SSML support. Apple provides a small set of inline commands, such as [[slnc]] to insert a pause. Expression is otherwise limited to choosing a voice and setting a rate. If you need emotion, style, or cloning, a local model or a production studio is a better fit.

What are the say command defaults?

The default voice is whatever is selected in System Settings, so name a voice explicitly in scripts. On the tested build the default file output was mono, 22050 Hz, 16-bit AIFF. The default rate follows the system speech setting, and -r overrides it in words per minute.

Does the say command work offline?

Yes. The synthesis happens on the Mac using installed system voices, so no network connection is required to generate or play audio. Voices that are not installed are downloaded through System Settings, which does need a connection at that moment.

Sources

Try Murmur - $49 one-time. No subscriptions, no cloud, no per-character fees.

Originally published at murmurtts.com

Camtasia Text to Speech: Local Alternative Guide

Tarun yadav — Sat, 07 Nov 2026 14:29:14 +0000

Camtasia text to speech runs inside a paid editor. See how to generate narration locally on a Mac and import it into Camtasia or any timeline editor.

Short answer: Camtasia text to speech lives inside Camtasia Audiate, TechSmith's companion app, and on the current plans the AI voiceover that generates narration is included only with the higher paid tiers. If you just need narration files to drop on a video timeline, you can generate speech locally on an Apple Silicon or Intel Mac with Murmur, export the audio, and import it as a normal audio track in Camtasia or any other editor. This guide explains how the built-in feature works, shows a local alternative step by step, and is honest about when the paid, in-editor option is still the better choice.

How Camtasia text to speech actually works

Camtasia's own text to speech is not a track type inside the Camtasia Editor timeline. It lives in Camtasia Audiate, and TechSmith states on its AI voice generator page that the feature is powered by ElevenLabs. The flow is short: you write or import a script into Audiate, pick a voice, adjust tone, speed, and dialect, then export the audio or send it into your Camtasia project. Audiate also lists a pronunciation dictionary and multi-voice narration, so you can assign different voices to different parts of one script.

The practical catch is plan gating. On the Camtasia Suite pricing page today, voiceover generation with 200+ voice options is marked as included only in the Create and Pro plans, not Starter or Essentials, and Audiate's multi-language support starts at Essentials. TechSmith sells these as annual subscriptions, and its store localizes currency by country, so check your own store page for the exact figure where you live rather than trusting a number from another region. One more thing worth separating: Camtasia's speech-to-text feature goes the other direction, transcribing existing audio for captions rather than generating narration.

When a local text to speech workflow fits better

If the goal is simply to get a clean narration track onto a video timeline, paying for a full suite subscription plus the AI voice tier every year can cost more than the job needs. A local text to speech app produces the same kind of audio file, and you import it as an ordinary clip. That keeps the editor you already like and separates the two purchases: the editor for editing, the voice tool for voice. Murmur fits that description. It is a local voice studio for Apple Silicon and Intel Macs running macOS 15 or later, it runs its models on the machine instead of uploading your script, and it exports audio you can import anywhere. It costs $49 one time, with a free plan and a 3-day Pro trial, and a 7 day refund policy.

Reach for a local narration workflow when most of the following are true for you:

You edit in Camtasia, ScreenFlow, Descript, DaVinci Resolve, Premiere, or Final Cut and only need narration files to place on a track.
You narrate on a steady schedule and would rather pay once than carry a per-year voice add-on.
Your scripts contain client names, unreleased products, or private research you would rather not upload to a cloud voice service.
You want the same voice across many videos, and you want it available on your machine rather than tied to a subscription tier.
You are comfortable doing one extra export and import step in exchange for owning the audio files.

Aspect	Camtasia Audiate AI voice	Murmur local audio, then import
Where it runs	Cloud voice service inside Audiate, powered by ElevenLabs	On your Mac (Apple Silicon or Intel)
Cost shape	Annual Camtasia subscription; voiceover generation needs Create or Pro	$49 one time, no per-character fees
Editor integration	Built into Camtasia and Audiate	Export a clip, then import into any timeline editor
Script handling	Script is processed by a third-party cloud voice service	Script stays on the machine
Voice options	Large stock library plus tone, speed, and dialect controls	Bundled and installable models, including cloning and voice design
Best for	One-tool shops that want narration generated inside the editor	Creators who want to own audio files and reuse them across tools

Step by step: local narration into a Camtasia style timeline

The workflow is the same in every timeline editor: generate, export, import, align. Here is the sequence that keeps re-edits cheap.

Write or paste the script into your text to speech app. Keep one paragraph per scene so a single line can be regenerated later without redoing the whole video.
Check pronunciation before you commit. Add tricky names, acronyms, and product terms to a replacement list, then listen. If you want a repeatable test, the local TTS pronunciation test suite is a good pattern to borrow.
Audition voices and pick one per role. If the video has an intro voice and a tutorial voice, generate both and name the files by role so you never mix them up.
Export each line or scene as a separate WAV file. WAV is the safest choice for editing because it is uncompressed and drops onto a timeline cleanly. The TTS audio export guide explains when to keep WAV and when to deliver MP3 or M4A.
Import the clips into the editor and place them on their own audio track above the screen recording. Camtasia style editors let you layer narration over system audio, so mute or duck the screen audio underneath the voice.
Trim and align. Cut the narration to match the on-screen action; do not cut the action to hurry the voice. If a line runs long, edit the script and regenerate that one line rather than speeding up the audio.
Export the finished video, then check captions against the final audio rather than the script draft.

If you are new to the audio side, the Mac text to speech workflow covers the setup, and local TTS for training videos covers narration pacing for how-to content. The same import idea applies to CapCut, Canva, and Clipchamp style editors, where you also bring in narration instead of typing it into a cloud generator.

Getting audio in and out cleanly

Two habits save the most time. First, keep one project folder with the script file and the exported clips together, so a re-edit is a re-export of one file and not a hunt through downloads. Second, use the same export settings for every clip in a project. Sample rate and channel layout should match across lines, because a single clip recorded at different settings can force the editor to resample and can shift sync slightly. Murmur exports files you can import directly, and it does not charge per character, so regenerating a line after a script tweak is a normal part of the loop rather than a fresh bill.

When you move data between tools, prefer a lossless format for the master and convert only for delivery. The audio export guide covers the formats in detail. Keeping a clean WAV master means your editor never starts from a compressed copy.

Where a local option is not the right call

Be honest about the tradeoffs. If you work on Windows, Murmur is not for you: it is a macOS app, while Camtasia itself runs on both Windows and macOS. If you need live speech while you record, a local batch text to speech app will not do it, because it generates files rather than acting as a real time microphone replacement. Murmur does not do speech to text either, so it will not transcribe or caption recordings you already have; for captions you still need a transcription tool. And if your team collaborates in the cloud with comments, shared projects, and browser access, a hosted editor suite fits that way of working better than a file you pass around.

There is also a convenience argument for the built-in route. Generating narration in the same window as the timeline, with no export and import step, is genuinely smoother. That convenience is real, and it is part of what the subscription buys. The local route trades that convenience for ownership of the files, no per-character cost, and keeping private scripts on your own machine. Pick based on which of those matters more for the video you are making.

Frequently asked questions

Is Camtasia text to speech free?

The AI voiceover feature is part of the paid plans. On TechSmith's Camtasia Suite pricing page, voiceover generation with 200+ voice options is included in the Create and Pro plans and excluded from Starter and Essentials. If you want narration without that subscription, generate the audio locally and import it.

Can I add narration to Camtasia without a voice subscription?

Yes. Generate the narration in a local text to speech app, export it as a WAV file, then import it into Camtasia and place it on an audio track above your screen recording. The editor treats it as an ordinary clip, so trimming, fading, and leveling work the same as any other audio.

Do I have to use Camtasia to use this workflow?

No. The generate, export, import, align steps work in any timeline editor that accepts standard audio files, including ScreenFlow, Descript, DaVinci Resolve, Premiere, and Final Cut. The editor does not need to know or care that the voice came from a local model.

Will a local voice sound the same as the cloud voices?

Do not assume that. Voice quality depends on the model and the settings, and it varies by voice and language. Murmur lets you choose between bundled and installable models, including cloning and voice design, so listen to the public samples and test the specific voice you plan to use before committing to a full script.

What Mac do I need for Murmur, and what does it cost?

Murmur runs on Apple Silicon and Intel Macs with macOS 15 or later. The website edition costs $49 one time, with a free plan and a 3-day Pro trial, and a 7 day refund policy. It does not do speech to text, so keep a separate tool for transcription and captions.

Sources

Try Murmur - $49 one-time. No subscriptions, no cloud, no per-character fees.

Originally published at murmurtts.com

Proofread by Listening: Text to Speech for Students

Tarun yadav — Fri, 06 Nov 2026 15:27:58 +0000

Proofread by listening with text to speech. A student workflow that catches dropped words, doubled words, and punctuation slips before you submit.

Short answer: To proofread by listening, paste your draft into a text to speech reader, set a slightly slow speed, and follow the words on screen while the voice reads them out loud. Listening pushes past the autocorrect your brain applies when you read your own writing, so dropped words, doubled words, and punctuation that changes meaning become easy to hear. You can start with a reader that is already on your device; a dedicated voice app is only worth paying for when you also want to produce finished audio.

The reason this works is a well-known quirk of reading. When you reread your own sentence, your eye often supplies what you intended to type instead of what is actually there. BBC Bitesize states it directly: your eye sometimes sees what should have been written rather than what is on the page, which is why reading the text aloud is their first proofreading method and why they point out that you can often 'hear' punctuation and grammar mistakes. Text to speech does the reading for you, so you can keep both eyes on the page and catch the mismatch between what the voice says and what the screen shows.

Why your eyes skim past your own errors

Seventy pages into a term, your brain is running a prediction engine. It knows what the argument is supposed to say, so it fills gaps and smooths over bumps without flagging them. That is efficient for comprehension and terrible for proofreading. Reading out loud breaks the pattern because speaking and hearing use a different path than silent reading, and that path does not get the same autocomplete.

Reading your own draft aloud does help, but it has limits. You get tired after a few pages, your pace drifts, and you are still the one deciding what the words are. A text to speech reader does not get tired, holds a steady pace, and never remembers what you meant. When the voice says a word that is not on your screen, or pauses somewhere a sentence should not pause, you have found a real problem instead of a hunch.

The other advantage is hands and eyes free for marking. You can keep a notebook, sticky flags, or a comment column open and annotate as you go, rather than holding a page and squinting. This turns proofreading from one slow pass into a set of short, focused passes, which is exactly how a full listen stays bearable.

The 15-minute proofread-by-listening workflow

Treat the audio like a checklist, not background noise. The sequence below gives you a repeatable pass that fits between finishing a draft and submitting it. Keep your original document open in one window and the reader in another, and fix the real document the moment you hear a problem.

Freeze the draft. Save a copy, then work from a listening copy so your editing never fights the text that is being read.
Clean the text for speech: expand abbreviations and symbols, break apart tables, and remove markup like footnote markers that will be read as noise. Keep the original formatting in the master file.
Pick a neutral voice and set the speed a little slow, roughly 0.9x. A natural, even voice keeps the focus on your words instead of the performance.
Pass one, missing and repeated words. Follow along word by word and flag anything the voice adds, skips, or doubles, such as a stray 'the the'.
Pass two, sentence shape. Mark run-on sentences, comma splices, and places where the meaning stalls because the punctuation does not match the thought.
Pass three, high-risk items. Slow down for names, dates, numbers, units, and citations, since those are where an error does the most damage.
Final pass, edges. Relisten to the first and last paragraph once more, because openings and closings are the lines graders and readers remember.

Fix in the original the moment you hear something, then resume listening. Do not rewind repeatedly to hunt; if a sentence keeps snagging, that repeated stumble is itself the signal that the sentence needs rewriting. Pausing is faster than rewinding, and it keeps your place tied to the visible text.

Which proofreading method to use when

Listening is the strongest single method for word-level and punctuation errors, but it is not the only one. Use this table to decide how to spend your last hour before a deadline, and combine methods rather than trusting any single pass.

Method	Best at catching	What it tends to miss	What you need
Silent reread	Big-picture gaps and obvious typos	Dropped words, doubled words, punctuation you meant to type	Your screen and attention
Read it aloud yourself	Punctuation and grammar that sounds wrong	Small typos at speed, and fatigue after a few pages	Your voice
Listen to text to speech and follow along	Missing words, repeats, run-ons, leftover placeholders	Homophones, citation style, and factual errors	A reader app or a built-in voice, plus headphones
Listen without following along	Overall rhythm, flow, and where a reader would get bored	Exact word-level errors, because your eyes are elsewhere	A reader app and full attention on the sound

What listening still will not catch

A voice that pronounces the words cannot tell you whether they are the right words. Homophones are the classic trap: their, there, and they're all sound identical, so a reader will never flag them. The same is true for to, too, and two, and for of versus off. These need a separate visual pass, ideally reading backward or one line at a time.

Listening also says nothing about citation format, reference-list styling, heading levels, or whether a quotation is exact. A voice will happily read a fabricated statistic or a misspelled author name at a confident, even pace. Keep a separate check for anything a grader will verify against a source, and confirm each fact and reference on its own.

One more caution: when a reader stumbles on a word, that stumble is not proof of an error. The engine may simply not know the name or the technical term. Check the spelling yourself before you change anything. For the related problem of engines that skip or clip words, the missing-word and punctuation troubleshooting guide explains how to separate a genuine text bug from a synthesis quirk.

Choosing a reader: built-in, browser, or studio app

You do not have to buy anything to proofread by ear. macOS ships with a spoken content feature you can trigger on selected text, and Microsoft Edge includes a Read aloud tool that narrates web pages, PDFs, and documents. Microsoft's own guidance names proofreading a paper as a primary use, and it lets you choose voices and change the reading speed, which is all most proofreading needs.

For students who need more than a quick reader, universities point to subscription tools. Yale's Student Accessibility Services lists Kurzweil 3000, Speechify, and NaturalReader as options, and notes that some are available to eligible students through the university, with free tiers that limit voice quality or reading time. If you want one of these to read a whole term of PDFs, check your school's accessibility office before paying, because the license is often already provided.

A production voice app is a different category of tool. Murmur is a local voice studio for Apple Silicon and Intel Macs running macOS 15 or later. It generates speech on your machine, so a private draft never leaves your laptop, and it charges no per-character fees. That matters if your proofreading copy includes unpublished work or personal notes. For plain read-aloud of a single essay, the free built-ins are usually enough; reach for a studio app when you also want consistent multi-voice audio or finished files you can replay on a walk. The one-time app versus subscription comparison walks through that trade-off.

Making a draft listenable without changing your words

A reader fails on drafts that were never meant to be spoken. Symbols, abbreviations, and broken line breaks confuse the voice and send you chasing fake errors. Before you listen, spend two minutes normalizing the copy: write out numbers and units, spell out symbols, and replace abbreviations such as e.g. and ibid. with plain words. This is preparation for the reader, not a rewrite, so do it on the listening copy and leave your submitted text alone.

The text normalization guide covers the common cases in more detail, including dates, ranges, and academic abbreviations. If you plan to keep a reusable script separate from your draft, the Mac text-to-speech workflow explains how to keep a clean master and a listening copy in step so a correction only has to be made once.

Access, privacy, and studying with audio

Text to speech is also an access tool, not only a proofreading trick. Yale describes it as software that produces audible versions of text in websites, ebooks, PDFs, and Word documents, and some students are approved for it as a formal accommodation. If reading a screen is tiring or difficult, listening lets you review material while walking, commuting, or resting your eyes, which can make a dense reading load more manageable.

Privacy is the other reason some students prefer a local setup. Drafts, feedback, and personal notes are exactly the kind of content you may not want on a server. Murmur runs models locally, so generating audio for a private document does not require an upload; the offline text to speech for private documents on a Mac guide explains when that matters and when a browser reader is fine. Two honest limits: Murmur is a Mac app, not a phone or Windows tool, and it converts text to speech only. It does not do the reverse, so it cannot transcribe a lecture or a voice memo.

If you are turning study notes into a listenable review track rather than checking a single essay, the same setup scales up. The course narration guide covers the writing and pacing side for longer educational audio, and the best text to speech apps for Mac roundup compares app-level options if you want to weigh readers side by side.

Frequently asked questions

Does listening to text to speech really help you catch errors?

Yes, for a specific class of errors. Hearing your words read back breaks the silent-reading autocomplete that hides dropped words, doubled words, run-on sentences, and punctuation that changes meaning. It will not catch homophones, citation style, or factual mistakes, so pair it with a short visual pass for those.

What speed should I use for proofreading by ear?

Start slightly slower than normal, around 0.9x. Slow speech makes missing and repeated words easier to hear without dragging on. Speed up only for a final rhythm and flow pass, when you are judging pacing rather than scanning for word-level errors.

Can I proofread a PDF by listening?

Often yes. Microsoft Edge Read aloud narrates PDFs, and many accessibility readers handle PDFs and Word documents. Scanned PDFs that are images need OCR first, otherwise the reader has no text to speak. Always follow along with the page so you catch formatting and layout problems a voice cannot report.

Is text to speech the same as speech to text?

No. Text to speech turns written words into audio, which is what you use for listening to a draft. Speech to text does the opposite and transcribes spoken words into writing. Proofreading by listening uses text to speech, so it needs text as the starting point.

Do I need a paid app to proofread by listening?

Usually not for a single paper. macOS has a built-in spoken content feature, and Microsoft Edge includes Read aloud for free. Paid readers add better voices, more file types, and unlimited reading, and your university may already license one. Only pay when you need those extras or want to produce finished audio rather than hear a quick read.

How do I handle names and numbers the voice reads wrong?

Slow down for those sections and check them against your source rather than trusting the voice. A reader mispronouncing a name is not the same as a typo. If you are generating audio you will reuse, spell out or annotate tricky items on the listening copy so the voice reads them the way you intend.

Sources

Try Murmur - $49 one-time. No subscriptions, no cloud, no per-character fees.

Originally published at murmurtts.com

Mac Text to Speech Voices: Download Enhanced and Premium

Tarun yadav — Fri, 06 Nov 2026 15:27:52 +0000

Download enhanced and premium voices on macOS, switch between them, and see when a local AI voice like Murmur beats Apple's built-in text to speech.

Short answer: Your Mac already speaks text out loud, and you can add larger, more natural voices by downloading them from Apple. Open System Settings, click Accessibility, click Read and Speak, then use the System voice menu and the info button to add a voice; any voice with a Download symbol is fetched from Apple and needs an internet connection once. The honest comparison is this: Apple's voices are free, private, and ready in seconds for read-aloud, while a local AI voice app is the tool you reach for when the job ends in an exported audio file, a cloned or designed voice, or a multi-speaker project.

This guide covers how to download and sample higher-quality macOS voices, how the quality tiers differ, how to fix pronunciation without changing voices, and where a built-in Mac voice stops being enough. If you only want to hear a document read back, you can stop after the steps below. If you are producing narration, audiobooks, or videos, the second half of the guide is the part that matters.

Where Mac text to speech voices live

Apple keeps every speech voice behind one settings pane. On macOS 15 and later you reach it from the Apple menu under System Settings, then Accessibility, then Read and Speak. Apple's own Change the voice your Mac uses to speak text page lists the same path: System Settings, Accessibility, Read and Speak, System voice. That page is also where Apple explains that you can change the language your Mac uses to speak, sample a voice before using it, and add new voices. Nothing in this workflow is hidden in a paid tier; the voices are part of macOS.

Read and Speak is the same system that powers the Option-Esc shortcut for speaking selected text. Apple's Have your Mac speak text that's on the screen page describes turning on Speak selection, pressing the shortcut, and using the onscreen controller to change rate, skip forward or back, pause, and stop. The voice you pick in Read and Speak is the voice that shortcut uses, so it is worth choosing deliberately rather than leaving the default.

How to download enhanced and premium voices on Mac

Follow these steps in order. The labels in the voice list can vary by macOS version and by language, so treat the names below as a map rather than exact wording on your machine.

Choose Apple menu > System Settings, then click Accessibility in the sidebar. You may need to scroll to see it.
Click Read and Speak.
Click the System voice pop-up menu and choose a voice, or click the info button next to System voice to browse the full list.
In the list, click the name of a voice to add it. If a Download symbol appears next to the name, macOS downloads it from Apple, so you need to be online for that step.
Click the Play symbol next to a name to sample the voice before you commit to the download.
To find a voice for another country or language, click the language in the sidebar, then click the name on the right. You may need to scroll down to see more voices for the current language.
When you are finished adding voices, click Back, then Done. A voice is not available until its download completes.

Two small habits save time. Sample first, because some voices are better suited to reading a full chapter than others, and a larger download that you never use is just wasted disk space. Second, note the name you picked. When you later build a reading workflow, the voice name is the one setting you may want to change per project; the rest of the setup stays the same.

Compact, Enhanced, and Premium: what the tiers mean

When you open the full voice list, the same first name can appear more than once with a quality label after it. The base voice is small and already installed so speech works immediately. The labeled versions are larger downloads that sound smoother and less synthesized. The exact set of tiers, and the size of each download, depends on your macOS version, the language, and the voice. Use the table as a general map, then check the list on your own Mac for the real options.

Voice tier	What it is	Download	Best for
Compact	The small voice that ships with macOS so speech works out of the box.	Already on your Mac	Quick read-aloud, short previews, and accessibility.
Enhanced	A larger download that sounds smoother than the compact voice.	Downloaded on demand; size varies by voice and language	Longer listening sessions when the base voice sounds too robotic.
Premium	The largest tier Apple offers for on-screen speech, downloaded on demand.	The biggest download of the three; size varies	The most natural read-aloud voice macOS can produce.

A practical test is to add the compact and the larger version of the same voice, then play the same sentence through each. Apple's guide points out that the Play button lets you sample a voice, which is the fastest way to decide whether the extra download is worth it. If the difference is small for your use, keep the compact voice and spend the space elsewhere.

Fix pronunciation before you change the voice

Names, acronyms, and product terms are usually the reason a Mac voice sounds wrong, not the voice itself. Apple's voice page describes a Pronunciations feature inside Read and Speak: turn it on, click the info button, and define a custom pronunciation. You can dictate or spell out how you want the word spoken. That is a targeted fix. Correcting one brand name is faster and cheaper than downloading three more voices hoping one handles it.

You can also adjust how a voice sounds, including how fast or loud it speaks, from the same voice settings. And you can remove a voice you no longer want: open Read and Speak, click the info button next to System voice, swipe left with two fingers on the voice, then click Delete. That is worth doing after you have tested a few larger voices and settled on one, because the downloaded tiers take up real space. For more on shaping a repeatable reading setup, see the Mac text to speech workflow.

When Apple's built-in voices are the right choice

Built-in voices win in three situations. First, listening: proofreading a draft, hearing an article, or getting through a long PDF without reading every line. Second, privacy: the voices run on your Mac and there is nothing to upload, which matters if the text is confidential; the offline text to speech guide goes deeper on that. Third, cost and speed: they are already installed and always available, with no account, no credits, and no per-character billing.

For a lot of people, that is the whole job, and paying for anything else would be wasteful. Be honest with yourself about what you actually need before you shop. If your end goal is to hear text, the right answer is usually the free one already sitting in System Settings.

Where Apple's voices stop being enough

Read and Speak is built for listening on screen, not for producing a finished audio asset. It does not hand you a project file, a timeline of multiple speakers, or a library of voices you designed. The moment your deliverable is an audio or video file, you are in production territory, and the tool that matters is one built around export.

You need a deliverable file, such as a WAV or MP3, rather than text spoken on screen.
You need a specific voice that is not in Apple's list, or a voice that is not public at all.
You need more than one speaker in the same piece, with consistent voices across edits.
You need to direct delivery, pacing, or style, and redo takes without re-downloading anything.
You need a language or accent that Apple's installed voice list does not cover on your Mac.

Mac voices vs AI voices: the real comparison

The useful question is not which sounds better in the abstract, because neither option wins everywhere. It is which one fits the job. Cloud AI voice services such as ElevenLabs and similar tools can produce striking results and offer cloned voices, but they typically run in the cloud, meter usage, and require an account. If per-character pricing or uploading your script is a problem for you, those are real tradeoffs; the ElevenLabs alternatives on Mac and the ElevenLabs pricing writeups cover that ground, and prices change, so check the vendor page before you buy.

A local AI voice app sits between the two extremes: the natural sound you want for produced audio, but running on your own machine so there is no upload and no per-character fee. Murmur is one of those, and it is macOS only. The table below frames the comparison by job rather than by brand.

Question	Apple's built-in Mac voices	A local AI voice app (Murmur)
Where it runs	On your Mac, installed with macOS	On your Mac, with models downloaded locally
What it costs	Included with macOS	Murmur is $49 one-time, a free plan and a 3-day Pro trial, 7-day refund policy
Main purpose	Speaking text on screen for reading and accessibility	Producing audio for projects, with export
Voice choice	Apple's fixed voice and language list	Depends on the installed model; includes cloning and voice design on supported models
Multiple speakers	Not the intended workflow	Multi-speaker projects are supported
Platform	Any supported Mac	Apple Silicon or Intel, macOS 15 or later

Where Murmur fits, honestly stated

Murmur is a local-first voice studio for Apple Silicon and Intel Macs. It runs speech models on your machine, so there are no per-character fees and no script upload. Kokoro ships in the app with more than ten built-in voices across its supported languages, and additional models can be installed for cloning, voice design, or broader language coverage; the exact model list lives in the app, so check it there rather than trusting a list in an article. Because so much depends on which model you install, two people using Murmur for different jobs may have quite different experiences.

Where Murmur is not the right choice: it is macOS only, with no Windows, iPhone, Android, or browser version; it does not do live or real-time streaming speech; and it does not do speech to text. If any of those is a hard requirement, a cloud service or a different tool is the better fit, and we would rather say so than have you buy the wrong thing. For the broader landscape, the best text to speech apps for Mac roundup and the best TTS voices for narration guide are useful next reads, and noise about one-time versus subscription pricing is covered in the one-time app versus subscription comparison.

A decision checklist you can run in two minutes

If you only need to hear text read back, stay with the built-in Mac voices and download one Enhanced or Premium voice you like.
If the text is sensitive, prefer anything that runs on device, which includes both Apple's voices and a local model app.
If you need a finished audio file, a multi-speaker timeline, or a voice you cannot find in Apple's list, look at a production tool rather than Read and Speak.
If you are on a Windows PC or need a phone app, cross Murmur off your list and look at cloud services.
If per-character fees or uploading scripts are dealbreakers, prioritize local execution over cloud voice services.
If you want someone else's voice, check the license and the person's consent before you clone anything.

One caution: Apple's own Personal Voice, described on its Create a Personal Voice on Mac page, is a synthesised voice that sounds like you, but Apple states it is for personal, non-commercial use, is available only on Apple silicon, and works with Live Speech and apps you approve. It is an accessibility feature, not a commercial narration tool. Do not assume it can stand in for a licensed voice for paid work.

Frequently asked questions

How do I download enhanced voices on my Mac?

Open System Settings, click Accessibility, then click Read and Speak. Click the info button next to System voice to open the full list, then click the name of the voice you want. If a Download symbol appears next to that name, macOS downloads the voice from Apple, so you need an internet connection for that step. The voice is available once the download finishes.

Are Mac text to speech voices free?

Yes. The voices are part of macOS, including the larger Enhanced and Premium tiers, and there is no separate purchase or subscription for them. The only cost is the disk space the downloads take. You can delete voices you no longer use from the same Read and Speak pane.

Are Mac voices as good as AI voices like ElevenLabs?

They serve different jobs. Mac voices are excellent for reading text on screen, are private, and cost nothing, but they are not built to export a finished audio file or to clone a specific voice. Cloud AI services and local AI voice apps are aimed at produced audio. Which sounds better depends on your ears and your material, so sample both before deciding.

Can I use Apple's Mac voices in a podcast or video?

The Read and Speak workflow is designed for listening on screen, not for producing a deliverable file, so it is not a natural fit for a project pipeline. If the output needs to be an audio or video asset, a tool built around export is usually the better choice. Check the terms that apply to your specific use and platform.

Do downloaded Mac voices need the internet to work?

No. Downloads require a connection once, but after a voice is installed it runs on your Mac without a network connection. That on-device behavior is why built-in voices are a reasonable choice for private or offline reading.

What is the difference between Compact, Enhanced, and Premium voices?

They are quality tiers of the same idea. Compact voices are small and installed with macOS so speech works immediately. Enhanced and Premium are larger downloads that generally sound smoother and less synthesized. The available tiers and download sizes vary by macOS version, language, and voice, so check the voice list on your own Mac.

Sources

Try Murmur - $49 one-time. No subscriptions, no cloud, no per-character fees.

Originally published at murmurtts.com

Narrated Presentation Script Template (With Timing)

Tarun yadav — Thu, 05 Nov 2026 15:53:02 +0000

A fill-in narrated presentation script template with per-slide timing and speaker notes, plus a words-to-seconds timing model and a worked ten-slide example.

Direct answer: a narrated presentation script template is one row per slide with four fields: the point on screen, the spoken narration, your private speaker notes, and a timing target in seconds. Fill the timing field by dividing the slide's spoken word count by your speaking pace, about 130 words per minute for a calm presentation, then round up for pauses. Set those same seconds as the slide's advance time so the deck moves on cue. This guide gives you the blank template, a words-to-seconds timing model, and one worked ten-slide example.

A narrated presentation is simply a deck with recorded voice, which means the script is the only part you can edit and regenerate without re-recording the whole thing. That is why a spreadsheet-style template beats a paragraph of prose: it keeps what the audience sees, what they hear, and what only you read in separate columns that can be revised independently. This structure suits training modules, course lectures, product walkthroughs, and any deck that will be recorded rather than delivered live. If the deck is a class, the course narration guide covers the writing side in more depth. If it is a workplace training deck, the training video narration guide covers the wider production loop from outline to export.

What a narrated presentation script template contains

Column	What goes in it	Rule of thumb
Slide	The slide number and the single idea on that slide.	One idea per row. If a slide carries two ideas, split it into two slides.
On-screen text	The shortest version of the point, for the slide itself.	Keep the visible text short. The narration carries the explanation.
Spoken narration	The exact words the audience hears.	Read the point, add the takeaway, stop. Two sentences on a content slide.
Speaker notes	Cues only you need: emphasis, a number, a short story, a reminder to pause.	Write these for a reader, not a listener. Never read them word for word.
Timing (seconds)	How long the slide stays on screen.	Spoken words divided by your pace, plus a pause for reading or a demo.

Copy those five fields before you write a single polished sentence. Fill the slide number and the one idea first, because every other column depends on it. Then draft the spoken narration, add a private note or two, and compute the timing last. The order matters: if you guess a timing before the narration exists, you will pad or trim the spoken words to hit a number that was never grounded in the content.

Time each slide before you record anything

Timing starts with word count, not with a stopwatch. VirtualSpeech's guide to speaking rate, which draws on the National Center for Voice and Speech, puts a comfortable presentation pace at roughly 100 to 150 words per minute, and notes that popular TED talks run faster, averaging about 173 words per minute. Use 130 words per minute as your planning default, since it leaves room for the pauses a recorded deck needs. The formula is simple: seconds = (spoken words / 130) x 60. A 90-word slide therefore runs about 42 seconds of speech before you add breathing room.

Slide type	Target spoken words	Seconds at 130 words per minute
Title	10 to 20	5 to 9
Agenda or roadmap	30 to 45	14 to 21
Content, one idea	70 to 100	32 to 46
Demo or screenshot	50 to 70	23 to 32
Section break	5 to 10	2 to 5
Summary or close	40 to 60	18 to 28

Treat the seconds column as the floor, not the ceiling. A slide that shows a chart, a diagram, or a list needs time for the audience to read it, and a slide with a demo needs time for the action to finish. Add two or three seconds to any slide the viewer has to read, and more if a click or animation has to play out. Round everything up to the nearest second so your advance times do not cut a word in half.

The word-count habit also catches a common error. A script you can read silently in twenty seconds can take twice that to say aloud, because reading speed and speaking speed are different. If a slide feels cramped, the fix is almost always to move a sentence to the next slide rather than to speed up the voice.

Speaker notes are not the spoken narration

Microsoft's help page on speaker notes describes the Notes pane as a box below each slide, visible on your monitor during a presentation but not to the audience. That makes the Notes pane the right place for a hint you deliver off the cuff, not a second copy of the narration. In a recorded deck the two roles split further: the spoken narration is what gets voiced and exported, while the notes are a script for the person presenting or for the editor assembling the video. Mixing them produces a recording that reads its own stage directions out loud.

Slide 4, Registration
Spoken narration: In this step, the user registers a device from the account page. Point out that the confirmation appears in the top right, and that no download is needed.
Speaker notes: Slow down here. Pause after confirmation appears. Say the word register clearly; do not rush the sentence about the download.

Slide 5, Email confirmation
Spoken narration: Before the account is active, the user confirms the email address. The link expires in one hour, and they can request a new one.
Speaker notes: If a tester asks, the expiry time is set by the account owner, not by the user.

Notice that the notes carry instructions and background that only the presenter needs, while the narration stands alone as something a listener could follow without seeing the slide. A quick test: cover the on-screen text and read only the spoken column. If the result is confusing, the narration is leaning on the slide instead of explaining it.

A worked example: a ten-slide training deck

Here is the same template applied to a short training module. The word counts are targets for planning, and the seconds are the word count divided by 130 words per minute. The total shows how a small number of words per slide adds up to a finished runtime.

Slide	Spoken words	Seconds at 130 words per minute	Speaker note cue
1 Title	15	7	State the topic and the one outcome. Then pause.
2 Agenda	40	19	Name the sections only. Do not explain them yet.
3 to 8 Content	540 (90 each)	249 (about 42 each)	One takeaway per slide. Pause after each point.
9 Demo	60	28	Narrate the click, then the result on screen.
10 Summary	50	24	Repeat the outcome and the single next step.
Total	705	about 325 seconds (5 minutes 25 seconds)	Add two to three seconds per slide for reading.

That total lines up with a familiar planning shortcut: a five-minute spoken piece usually needs roughly 650 to 800 words at a normal pace. Because the template tracks words per slide, you can see immediately which slide is doing too much work. In this example the content slides carry most of the weight, which is correct for training, while the demo stays short so the action can breathe. If the deck ran long, trimming one content slide would save about 42 seconds without touching the voice speed.

Decide what belongs in the script and what stays in your notes

Put anything the audience must hear in the spoken narration, and nothing else.
Put reminders, pronunciation notes, and emphasis cues in the speaker notes.
Spell out numbers and symbols the way you want them voiced (write five million, not 5M).
Keep a name or term that the voice could mispronounce next to the narration as a note.
Cut transitions that only make sense when a cursor is moving, such as as you can see here.
Mark each slide that needs a longer pause, so it is deliberate rather than accidental.

The checklist is a sorting tool, not a style rule. A note that you find yourself reading out loud on every pass has quietly become narration, and should move into the spoken column. A spoken sentence that no listener could follow without seeing the slide belongs in the notes or, more often, needs to be rewritten until it stands on its own.

Turn the script into per-slide audio

Once the spoken column is final, generate one audio file per slide and name each file after its slide number, so a later revision is traceable to the right place. Murmur is a local-first macOS voice studio for Apple Silicon and Intel Macs running macOS 15 or later. It handles text to speech, voice cloning, voice design, and multi-speaker projects, and it runs the models on your own machine, so there are no per-character fees and nothing is uploaded to a cloud service. That matters for a training deck that contains unpublished product names, internal process details, or anything under an NDA. Pick one voice and keep it for the whole deck: a voice that changes between slides reads as a mistake even when the words are identical. The Mac text-to-speech workflow walks through the script-to-audio handoff, and the audio export guide explains the difference between an editing master and a delivery file. Murmur costs $49 one-time and has a free plan and a 3-day Pro trial.

If you are still choosing a voice for the narration, the narration voice roundup compares the options. Whichever tool you use, the timing column you filled in earlier becomes the slide's advance time, so the deck moves on the schedule you planned instead of waiting on a fixed interval.

When Murmur is not the right choice

Murmur is a Mac app, so if you work on Windows, need live or real-time streaming speech, or want several people editing one project in the cloud at once, it is the wrong tool and a cloud service with those features will serve you better. It also does not do speech to text, so it cannot turn a recorded interview into a script. And if a slide needs a specific celebrity voice or an emotion that a stock generated voice cannot carry, plan on recording a human narrator instead. Murmur is strongest when you want private, repeatable narration that you can regenerate yourself from the same script, slide after slide.

Frequently asked questions

What is a narrated presentation script template?

It is a fill-in worksheet with one row per slide. Each row records the slide's single idea, the on-screen text, the spoken narration the audience will hear, the private speaker notes the presenter reads, and a timing target in seconds. Keeping those fields separate lets you revise the narration without disturbing the notes or the visible slide.

How do I time each slide?

Count the spoken words on the slide and divide by your speaking pace, then multiply by 60 to get seconds. At about 130 words per minute, a 90-word slide runs roughly 42 seconds. Round up and add two to three seconds for any slide the audience has to read, and more for a demo. Use that number as the slide's advance time.

What is the difference between speaker notes and narration?

Speaker notes are cues only the presenter sees, such as emphasis, a reminder to pause, or background detail. Narration is the exact text the audience hears and the text that gets recorded. Microsoft's help page notes that the Notes pane is visible on your monitor but not to the audience, which is why a note should never be read aloud word for word.

How many words is a five-minute narrated presentation?

Roughly 650 to 800 spoken words at a comfortable pace, which works out to about 130 to 160 words per minute. Because the template counts words per slide, you can total the spoken column and see whether the deck fits the time before you record anything.

Can I use this template for a recorded deck instead of a live talk?

Yes, and it helps most there. A recorded deck only needs the spoken column voiced, so you can generate one audio file per slide and regenerate a single slide when a fact changes. The timing column doubles as the slide advance time, so the recording and the visuals stay in step.

Should the narration repeat the text on the slide?

No. Keep the on-screen text short, and let the narration say what the slide means and why it matters. If the spoken column only reads the visible text back to the audience, the audience has no reason to listen, and the slide is doing the talking.

Sources

Try Murmur - $49 one-time. No subscriptions, no cloud, no per-character fees.

Originally published at murmurtts.com

E-reader Text to Speech vs Making Your Own Audio

Tarun yadav — Thu, 05 Nov 2026 15:52:57 +0000

Compare e-reader read-aloud with generating your own chaptered audio on a Mac, and see when to listen now versus when to own a finished file.

Direct answer: if you want to listen to a book you already own, use the read-aloud feature that is already inside your reader or reading app, such as Kindle's Assistive Reader or a book on a Mac. If you want a file you own that plays on any device, with one consistent narrator and real chapter markers, you generate the audio yourself from the book's text. The two paths answer different questions: listen to what you have right now, or produce a finished audio file you keep.

This guide separates those two jobs. It explains what an e-reader with text to speech does well, where its limits show up in practice, and how to turn an EPUB or PDF into your own narrated audio on an Apple Silicon or Intel Mac when a built-in voice is not enough. It is about listening to books and documents you own, not about buying audiobooks from a store.

What e-reader text to speech actually does

A modern e-reader or reading app can read a book aloud, and it usually does this with the operating system's text-to-speech voices rather than a narrator recorded for that title. On Kindle, the feature is called Assistive Reader. Amazon's own page for Assistive Reader on Kindle E-Reader devices states that it is available on Kindle E-Reader 11th Generation and higher, Kindle Scribe, and Kindle Colorsoft, that it underlines the text as it is read, and that you can play, pause, change the reading speed, or rewind by 30 seconds. It also states that the feature supports books in English, Spanish, French, German, and Italian, and works only with books that have Enhanced Typesetting.

The Kindle app behaves differently from a dedicated device. Amazon's Assistive Reader in Kindle apps page says the app version runs on iPhone, iPad, Android, Mac, and Fire tablets, uses the device's operating system text-to-speech engine to read the book, and supports books purchased from the Kindle store, borrowed through Kindle Unlimited or Prime Reading, or shared through a Family Library. That detail matters: the voice you hear often comes from your device's speech settings rather than from one studio recording shipped with the book.

Apple's route is split in two. Books that ship with a Read Aloud feature play a narrator's audio with synchronized highlighting, but Apple's Books Asset Guide states that read aloud content is supported only in fixed-layout books, and that the text and audio are synced using Media Overlays, a SMIL subset of EPUB. For every other book, the Mac can instead speak the text. Books on Mac lets you choose Edit, then Speech, then Start Speaking, and the system-wide Read and Speak settings described on Apple's Have your Mac speak text that is on the screen page let you set a shortcut (Option-Esc by default), highlight words or sentences as they are spoken, and show a small onscreen controller.

Where built-in read-aloud runs into limits

Built-in read-aloud is convenient, but it is tied to conditions you do not control. The book has to be eligible (Enhanced Typesetting on Kindle), the language has to be supported by the feature or by the device's speech engine, and the voice is whatever the device provides, which Amazon's own comparison notes is often one voice per language on Kindle devices. Some Kindle e-readers also need a paired Bluetooth speaker or headset because they have no speakers. Most importantly, the audio is a live rendering inside the app: normally there is no separate audio file you can copy to a phone, drop into a video edit, or hand to someone else.

Eligibility: the title has to support the feature, for example a Kindle book with Enhanced Typesetting enabled.
Language: both the book text and the read-aloud voice need to be supported by the feature or the device's speech engine.
Voice choice: on Kindle devices there is typically one voice per language, while voice selection on the apps comes from the device's own text-to-speech settings.
Hardware: many Kindle e-readers need a paired Bluetooth device for audio output.
Output: the speech stays inside the reader as playback, so there is usually no delivered file to keep or move.

When built-in read-aloud is the right tool

Choose the reader's own voice when the goal is simply to listen to a book you already have, especially a single title in a supported language. It is included with the reader, it needs no setup, it remembers your place, and it covers the whole book without you converting anything. If you want to follow along with synchronized highlighting for readability or accessibility, that highlighting is a genuine advantage a plain audio file cannot reproduce. For casual listening, the built-in feature is usually the smarter choice, and reaching for a production tool would be unnecessary work.

When to generate your own audio instead

Reach for your own generation when the audio itself is the deliverable, not just the act of listening. That covers an unpublished manuscript you want to hear and keep, a set of documents you want served in one narrator's voice, or a book you want as a chaptered file that plays offline on any device, can be edited, and can be reused. It also matters when a title is not eligible for read-aloud, when the language you need is not supported by the reader, or when you want more than one voice, for example a distinct narrator plus separate dialogue voices.

This is the job Murmur is built for. Murmur is a local-first macOS voice studio for Apple Silicon and Intel Macs running macOS 15 or later. It imports a PDF or an EPUB, extracts the text section by section in reading order, and runs the speech models on your own machine, so there are no per-character fees and nothing is uploaded to a cloud service. That matters for a private manuscript or a client document. The offline text to speech for private documents guide covers the privacy side, and the audiobook narration guide covers choosing and auditioning a narrator for a long book.

Decision aid: match the job to the tool

Your goal	E-reader read-aloud	Generate your own audio
Listen to a purchased book now	Yes. Built in, included with the reader, remembers your place.	Works, but you import and generate first.
Follow along with synchronized highlighting	Yes. The reader underlines or highlights as it speaks.	Not in a standalone audio file.
Own a file that plays anywhere offline	Usually no separate file is produced.	Yes. Export WAV or M4A and keep the file.
One consistent narrator across many documents	Depends on the device voice, often one per language.	Yes. Reuse one narrator, or clone or design a voice.
Multiple voices, for example narrator and characters	Rare in read-aloud features.	Yes, with multi-speaker projects.
Turn a manuscript, PDF, or EPUB you wrote into audio	Only if the title is eligible and the language is supported.	Yes. Import the document and generate the audio.

Treat the table as a shortcut, not a rule. Many people use both paths: the reader's voice for a quick listen, and a generated file when the audio needs to leave the app. Deciding early which one you actually need saves you from either doing unnecessary production work or discovering too late that the audio you wanted to keep does not exist as a file.

How to turn an ebook into your own narrated audio on a Mac

The steps below assume you own the text and have the right to make an audio version of it. Recreating a copyrighted book as audio for distribution is a licensing question, not a technical one, and it is separate from what any tool can do. For personal listening of something you wrote or have permission to use, the process is straightforward.

Get the text into a file you can read: a PDF, an EPUB, or plain text. Murmur imports PDF and EPUB through its file importer and accepts plain text and .text files in the composer.
Check text quality early. Open the file and read one page of body text. Scanned or locked documents fail because there is nothing to extract, so fix the source before you generate anything.
Let the tool split the document. Murmur extracts text in reading order and breaks it into chunks of roughly 400 to 1,000 characters, aiming for about 500, splitting at paragraph and sentence ends and keeping a section or page number with each chunk.
Group chunks into one script per chapter, using the table of contents or headings as boundaries. One script per chapter means a later correction touches one chapter instead of the whole book.
Choose one narrator and use it for every chapter so the book sounds continuous.
Generate, then listen at the start and end of each chapter. Spot check names, numbers, and any line the voice mispronounces, and fix the script rather than trying to edit the audio.
Export a master and a delivery file. Murmur exports WAV (a lossless editing master) and M4A, and writes timeline markers into the M4A as chapter markers, with optional transcript and chapter list files.
Keep the scripts next to the audio, so a future change can be regenerated in one place.

These steps are the same whether the source is an ebook or a stack of PDFs; only the extraction differs. If the document has a clean text layer and a clear chapter structure, the work is mostly choosing boundaries and proofing names. If the source is really a web article or a set of posts rather than an ebook, the process is closer to turning blog posts into audio, which has no locked-file problem at all.

A checklist before you generate a whole book

Confirm you have the text and the right to produce an audio version of it.
Check that the document opens and the body text is selectable, not an image.
Confirm the chapter boundaries you plan to use.
Write down how names, numbers, and abbreviations should be spoken.
Pick one narrator and audition it on a hard page with dialogue, names, and dates.
Generate and review one chapter fully before queuing the rest.
Decide your master format and your delivery format up front.
Keep scripts, audio, and a chapter list together in one project.
Listen to every chapter boundary once, because joins are where problems hide.

The WAV, M4A, and MP3 export guide explains why you keep a lossless master and deliver a smaller file with chapters. The AI audiobook workflow goes deeper on pacing and structure across many chapters. Both assume the text is already clean, which is exactly why the text check comes first in this guide.

When Murmur is not the right choice

Murmur runs only on Macs (Apple Silicon or Intel) with macOS 15 or later. It is not a Windows, iPhone, Android, or browser app, it does not provide live or real-time streaming speech, and it does not do speech to text. If you just want to hear a book you already bought, the reader's built-in read-aloud is simpler and you should use it. If you need several people editing one project in the cloud at once, a cloud service fits that better. If a publisher requires a signed audiobook file with digital rights management, you will need a dedicated production tool on top of the audio you generate. Murmur is strongest when you want private, repeatable narration you can regenerate chapter by chapter, and when owning the file matters more than pressing play inside an app.

Frequently asked questions

Is it legal to generate audio from an ebook?

That is a rights question, not a tool question. Making a personal audio copy of a book you own may be allowed in some places and not others, and distributing or selling a generated audiobook of a work you do not have rights to is a different matter entirely. Check the license for the specific title and your local rules before you publish anything you did not write.

Can Kindle read any book aloud?

No. Amazon states that Assistive Reader works only with books that have Enhanced Typesetting and supports books in a set list of languages. A book without that setting, or in a language the feature does not cover, will not read aloud this way. That is one reason to generate your own audio when a specific title is not eligible.

Does Apple Books read every book aloud?

Not with a recorded narrator. Apple's Books Asset Guide says read aloud content is supported only in fixed-layout books that include it. For other books, Books on Mac can fall back on the system voice through Edit, then Speech, then Start Speaking, which reads the onscreen text rather than playing a produced narration.

What file do I get from e-reader read-aloud?

Typically none. The read-aloud is playback inside the app, so there is usually no separate audio file you can copy elsewhere. Generating your own audio is the way to end up with a file, and Murmur exports WAV and M4A, with chapter markers written into the M4A.

How do I turn an EPUB into an audiobook on a Mac?

Import the EPUB, let the tool extract and chunk the text in reading order, group the chunks into one script per chapter, choose a single narrator, then export. Murmur follows that sequence and produces a WAV master plus an M4A with chapter markers, so the result plays like a normal audiobook.

Sources

Try Murmur - $49 one-time. No subscriptions, no cloud, no per-character fees.

Originally published at murmurtts.com

Scripting voiceover on a Mac from the command line

Tarun yadav — Wed, 04 Nov 2026 17:30:00 +0000

The first version of a video script is never the last. A sentence gets shorter. A claim that sounded good in the outline gets cut because the footage does not support it. A scene moves. If the narration lives in a web app, every edit means opening a browser, pasting the new text, waiting, downloading another file, and dragging it into the timeline. Do that a few times and the voiceover stops feeling like part of the project. It becomes a manual step you avoid.

Treat narration like build output instead. Keep the script in the repo. Regenerate only the scenes that changed. A Mac can do this today, and the command line is the right place to wire it up.

Start with say, then notice where it stops

macOS ships with say. It is the zero-install baseline, and it is genuinely useful for checking timing before you commit to a voice.

say -o audio/intro.aiff "Rough timing pass."

For placeholder tracks, that is enough. For narration that ships, it is not. The built-in voices sound robotic, and say writes AIFF, so you usually need a conversion step before the file fits an editor. There is no manifest and no batch job. I would not build the final pipeline around it.

Murmur is a native macOS app for local text to speech. It bundles Kokoro, an open weight TTS model with 82 million parameters under Apache 2.0. The public model card is on Hugging Face. Murmur runs on Apple Silicon Macs with macOS 15 or newer. After models are downloaded, generation works offline, and scripts never leave the machine.

Install the CLI and check the basics

The CLI is a Pro feature. Open Settings in Murmur, then click Install CLI. That links ~/.local/bin/murmur to a signed helper inside the app, so keep ~/.local/bin on your PATH. The first checks are simple:

murmur status
murmur status --json
murmur models
murmur voices

murmur status --json is the one a script or agent should call. Readiness can be checked without parsing a human sentence. murmur models shows what is installed, and murmur voices gives you the voice IDs for a manifest.

The CLI and MCP setup is documented at Murmur's automation page.

Generation itself is one command:

murmur generate --text "In this scene, the narrator introduces the problem." \
  --voice af_bella --speed 0.95 --output audio/intro.wav

Or point at a text file:

murmur generate --input scripts/chapter-01.txt --model kokoro \
  --output audio/chapter-01.m4a

The input is UTF-8 text or Markdown. The output is WAV or M4A. Change the text, run the command, get a new file. No browser tab, no download to rename.

Put the scene list in the project

A shell loop works until the project has many scenes. At that point, a manifest is easier to review and diff. Commit it beside the video:

{
  "items": [
    {
      "id": "intro",
      "inputPath": "../scripts/intro.txt",
      "outputPath": "../audio/intro.wav",
      "voiceID": "af_bella",
      "speed": 0.95
    },
    {
      "id": "chapter-01",
      "inputPath": "../scripts/chapter-01.txt",
      "outputPath": "../audio/chapter-01.m4a",
      "voiceID": "af_bella",
      "speed": 1.0
    }
  ]
}

Relative paths resolve from the manifest's folder, so ../scripts/intro.txt points outside video/. This keeps the manifest portable with the project as long as the layout stays consistent. Then run:

murmur batch --manifest video/scenes.json

Batch work is long enough that the CLI may hand back a job ID. Use murmur jobs get ID to check it and murmur jobs cancel ID to stop it.

A Makefile for changed scenes only

A manifest is the right shape for a full render. For day-to-day edits, a small Makefile that touches only changed scenes is better. It assumes one text file per scene and writes one WAV per scene:

SCENES := intro chapter-01 chapter-02
AUDIO := $(addprefix audio/,$(addsuffix .wav,$(SCENES)))

VOICE := af_bella
SPEED := 0.95

.PHONY: all clean

all: $(AUDIO)

audio/%.wav: scripts/%.txt
    @mkdir -p $(@D)
    murmur generate --input $< --voice $(VOICE) --speed $(SPEED) --output $@

clean:
    rm -f $(AUDIO)

Run make after changing a script. Make compares timestamps and runs the rule only when scripts/scene.txt is newer than audio/scene.wav. That is the useful part. Narration stops being a separate task and becomes another build artifact.

Make tracks files, not variables. If you change VOICE or SPEED, existing WAVs are still up to date as far as it knows. Run make -B or make clean when those values change.

Let a coding agent run it

murmur mcp serve starts a stdio MCP server. One registration example is:

claude mcp add --transport stdio --scope user murmur -- <helper path> mcp serve

Replace <helper path> with the signed helper inside the app. Once the server is connected, an agent can start generation and poll murmur_get_job for long operations. That turns a small script edit into a job the agent can run, without waiting on you.

What it costs and where it stops

The CLI and MCP server are Pro features, and both are Mac only. The local generation model needs disk space and memory. The free plan covers Kokoro voices with a 1,500 character limit per generation. There is a 3-day Pro trial. Pro is $49 one-time, with no subscription and no account required. If it does not fit, the refund window is 7 days.

Other limits matter depending on the work. There are no cloud voices. Cloning quality is below ElevenLabs. Cloning also requires an optional local bridge and at least 16 GB of memory. There is no team admin UI. If you need browser access from any machine, Murmur is the wrong tool.

Where ElevenLabs is the better choice

ElevenLabs has the stronger voice product. Its cloning and expressive voices are best in class, its multilingual catalogue is large, and it works in any browser on any OS. It also has an API and team features, plus dubbing. If you need the best clone, a large language list, or shared team controls, use ElevenLabs. You can compare their plans on the ElevenLabs pricing page.

Disclosure: I build Murmur.

The setup is small: one text file per scene, a Makefile with timestamps, and a manifest for full runs. If you want to try it, start here: Murmur. The 3-day Pro trial includes the Pro features, so you can test the CLI and MCP path before paying.

Final Cut Pro AI Voiceover Workflow (Roles, Sync, Retime)

Tarun yadav — Wed, 04 Nov 2026 15:50:58 +0000

Set up an AI voiceover in Final Cut Pro: assign audio roles, sync narration to picture, retime clips without pitch drift, and export clean stems.

Direct answer: to add an AI voiceover in Final Cut Pro, generate one audio file per section on your Mac, import those files into a library event, give each clip the Dialogue role, place them on the timeline, sync them to picture, and retime any clip from the Retime menu with Preserve Pitch left on. Final Cut Pro treats narration as ordinary audio, so the real work is roles, sync, and retiming, not any built-in AI feature. Nothing in the app writes speech; it edits audio you bring in.

That is the whole shape of the job, and it is why the workflow begins before you open Final Cut Pro. You write a script, generate the clips elsewhere, then use the editor to place, label, and time them. The steps below assume a scripted voiceover for a talking-head, a screen recording, or a training video. If your narration is one continuous read, the same ideas apply with fewer and longer files. For the wider writing loop, the Mac text-to-speech workflow covers how the script and the audio stay in step.

Roles decide how your voiceover behaves

Roles are metadata labels that identify what a clip is. Apple groups them into five default categories, Video, Titles, Dialogue, Music, and Effects, and Final Cut Pro assigns one automatically when you import a clip based on its existing metadata. Three more roles, Adjustment Clips, Captions, and Subtitles, appear when you create or import those item types. Roles are color coded, and they are the reason the timeline stays readable once a project has dialogue, music, and sound design at the same time. If the role assignment is wrong, every later step, from mixing to export, gets harder.

Role	What belongs there	What it means for a voiceover
Dialogue	Spoken content: narration, interview audio, voiceover	Your generated narration belongs here. It stays separate from music and can be exported on its own.
Music	Songs, score, and beds	Anything that plays under the voice. Keeping it in its own role makes it easy to duck later.
Effects	Sound effects and Foley	Impacts, whooshes, and room tone. This fills the gaps between dialogue rather than covering it.
Video and Titles	Picture and on-screen text	Separate from audio, but they share the same color coding system in the timeline.

The role also controls delivery. Apple documents exporting a separate file, called a media stem, for each role, either as a combined multitrack QuickTime file or as separate audio and video files. You can set mono, stereo, or surround output for the audio channels during that export. That is the practical payoff: when narration lives in the Dialogue role, you can hand a clean voice stem to a sound mixer, or re-export just the voice, without touching the music bed.

Generate the narration before you touch the timeline

Generate one file per section rather than one giant file. Short clips are easier to sync, easier to retime, and cheaper to redo when a single line changes. A section here means a paragraph, a slide, or one screen of a walkthrough, not a whole chapter. Murmur is a local-first macOS voice studio for Apple Silicon and Intel Macs running macOS 15 or later. It handles text to speech, voice cloning, voice design, and multi-speaker projects, and it runs the models on your own machine, so there are no per-character fees and no cloud upload. That matters when the video contains an unreleased product, a client script, or anything under an NDA. Murmur exports WAV and M4A audio, and WAV is one of the audio formats Apple lists as supported in Final Cut Pro. It costs $49 one time and has a free plan and a 3-day Pro trial.

Write the script in sections, one block per clip, so a wording change costs one regenerated file.
Pick one voice and one model for the whole video; switching mid-project is immediately audible.
Generate one audio file per section and name each file after its position or timecode.
Listen to a single sample before generating the rest, and fix pronunciation issues in the script.
Export WAV for editing when you want a lossless master to retime and re-export.

Naming is the part people skip. A file called intro-00-00.wav or step-04-register-device.wav tells you where it goes without opening it. Final Cut Pro also reads metadata on import, so a file whose tags say it is music can arrive already tagged as Music. Renaming and re-tagging before import saves a cleanup pass. A small quote file showing the section wording set against the file names can be enough to keep the two in sync:

00:00 intro.wav
Welcome to the walkthrough. In the next three minutes you will publish your first page.

00:14 step-01-open-dashboard.wav
Open the dashboard from the top left. The overview loads in a second or two.

00:31 step-02-create-project.wav
Choose New project, then name it after the client. Nothing is committed until you save.

If the same project also includes screen-recorded walkthroughs, the training video narration guide covers the writing and capture side end to end.

Import the clips and assign the Dialogue role

Once the files are in a library event, drag the first clip onto the timeline and check its role before you place the rest. Apple documents two reliable ways to change a role, and either works for a batch of narration clips.

Select the clip in the browser or the timeline.
Open the Inspector with Window > Show in Workspace > Inspector (Command-4), then click the Audio button at the top.
Scroll to the Audio Configuration section and use the Roles pop-up menu on the audio component to choose Dialogue.
Or skip the Inspector entirely: control-click the selection, choose Assign Audio Roles, and pick Dialogue from the submenu.

Do this for every narration clip. If a file has more than one audio component, expand the audio and label each component, which matters when a single file carries a mix. Everything added from the built-in Sound Effects category defaults to the Effects role, so if you pull a sting or ambience from there, leave it in Effects and keep Dialogue for the voice. Assigning roles early is what makes the final export a one-click job instead of an afternoon of soloing clips by hand.

Sync the narration to picture

Final Cut Pro has a real sync tool. With clips selected, choose Clip > Synchronize Clips (Option-Command-G), or control-click and choose Synchronize Clips. Apple states that the app analyzes the selection for sync points such as markers you have added, timecode, file creation date, and audio content. If it finds no shared sync point, it aligns the clips at their respective starting points. The option Use audio for synchronization makes a precision match using waveforms.

Select the video clip and its narration clip in the browser.
Choose Clip > Synchronize Clips (Option-Command-G).
Type a name for the new clip in the Synchronized Clip Name field.
Turn on Use audio for synchronization when the audio genuinely matches the picture.
Click OK. Final Cut Pro creates a new synced clip and leaves the original clips untouched.

Here is the catch specific to generated narration. Audio sync matches waveforms. A generated voice has no waveform relationship to the picture, so Use audio for synchronization cannot line it up the way it lines up a lav track against a camera scratch track. For AI voiceover, sync by timecode or markers instead. Put a marker on the frame where a section should begin, name the generated file to match, and align the clip start to that marker by hand. That is slower than one command, but it is the method that actually works when the voice was never in the room. Do not narrate a promise of frame-perfect sync that the tool cannot deliver for synthetic audio.

Retime clips without breaking the read

Retiming is where a scripted voiceover usually needs a nudge. Your picture may run a little long or short against the read, and Final Cut Pro can stretch or squeeze the audio to fit. Select the clip, then click the Retime pop-up menu below the viewer to reach Slow, Fast, Custom, and the retime editor. Apple notes that constant speed changes alter a clip's duration: setting a 5-second clip to 50 percent speed makes it 10 seconds long and plays it more slowly.

Retime option	What it does	When to use it on narration
Slow or Fast preset	Applies a uniform percentage speed change to the selection	Small trims to fit a line to picture. Keep changes subtle so speech does not drag.
Custom speed	Lets you type an exact playback percentage	When you already know the exact stretch needed to match a timing.
Show Retime Editor	Drag the retiming handle for a custom constant speed	Quick visual adjustment against the timeline, via Command-R.
Smooth Slo-Mo	Machine-learning slow motion on Apple silicon Macs	Visual slow-motion shots. Not for narration audio.

Apple states that Final Cut Pro preserves pitch by default when you retime a clip, so the voice does not sound higher when it speeds up or lower when it slows down. You can toggle this from the Retime pop-up menu: choosing Preserve Pitch shows a checkmark when it is on, and choosing it again turns it off. Leave it on for dialogue. Turning it off makes the voice chirp or growl, which is only useful as a deliberate effect. You can change this before or after a clip has been retimed.

Two housekeeping notes. Speed settings apply to the instance of the clip in your project, not to the source file on disk, so the original stays clean. If you actually need a baked file, export the clip as a QuickTime movie. And go gently: a few percent is invisible, but a large stretch makes even a good generated voice sound processed, so prefer fixing the script or the edit over a heavy speed change.

Export clean audio stems with roles

When the cut is locked, roles let you split the delivery. Apple's guide describes exporting a separate media stem for each role, either as a combined multitrack QuickTime file or as separate audio or video files, and you can assign mono, stereo, or surround output for the audio channels during export. For a voiceover job, that means a Dialogue stem of narration and a Music stem of the bed, delivered separately or together depending on what the client asked for.

Test the export before you rely on it, and confirm the sample rate and channel layout match the delivery spec. If you are deciding on formats, the WAV, M4A, and MP3 export guide explains master files versus delivery copies. And if you are still weighing how a dedicated editor compares with app-based voiceover tools, the Canva, CapCut, and Clipchamp voiceover comparison puts those options beside a full editing timeline.

When Murmur is not the right choice

Murmur is a Mac app, so if you work on Windows it is not an option. It also does not do live or real-time streaming speech, and it does not offer cloud collaboration where several people edit one project at once. It does not do speech to text, so it cannot transcribe an interview into a script. And if you need a specific celebrity voice, or an emotional performance a stock generated voice cannot carry, plan on recording a human narrator. Even inside Final Cut Pro, if the read is one long documentary take with heavy performance, a real voice through a real microphone will still beat a generated one. Murmur is strongest when the narration is scripted, private, and likely to change more than once.

Frequently asked questions

How do I add an AI voiceover in Final Cut Pro?

Generate one audio file per section outside Final Cut Pro, import the files into a library event, and drag them to the timeline. Then set each clip's role to Dialogue, sync it to picture, and retime it if the timing is off. Final Cut Pro edits the audio; it does not generate speech itself.

Why does audio sync not line up my generated voiceover?

The Use audio for synchronization option matches waveforms. A generated voice has no waveform relationship to your camera audio, so there is nothing for the tool to match. Sync generated narration by timecode or by adding a marker on the frame where each section starts, then align the clip by hand.

Does retiming a voiceover change the pitch in Final Cut Pro?

Not by default. Apple states that Final Cut Pro preserves pitch when retiming clips, so the voice does not sound higher when it speeds up or lower when it slows down. You can turn Preserve Pitch on or off from the Retime pop-up menu below the viewer, before or after a clip is retimed.

What role should a voiceover use in Final Cut Pro?

Use the Dialogue role. It is one of the default roles Final Cut Pro assigns automatically, and keeping narration in Dialogue, with music in Music and sound design in Effects, keeps the timeline readable and lets you export a clean voice stem later.

What audio format should I export from the narration tool?

WAV is the safe choice. Apple lists WAV among the audio formats supported in Final Cut Pro, and a lossless master avoids generation loss if you retime or re-export. Keep the original generated files so you can rebuild a clip without regenerating it.

Sources

Try Murmur - $49 one-time. No subscriptions, no cloud, no per-character fees.

Originally published at murmurtts.com

How to Do a Voiceover in DaVinci Resolve

Tarun yadav — Wed, 04 Nov 2026 15:50:52 +0000

Add a voiceover in DaVinci Resolve: record live or import generated narration, then clean it up in Fairlight with EQ, dynamics, noise reduction and ducking.

Direct answer: to do a voiceover in DaVinci Resolve you have two paths. Record it live on the Fairlight page by patching a microphone to a track, arming that track, and pressing record. Or generate the narration elsewhere, drop the audio file into the media pool, and place it on its own audio track in the timeline. Importing is the better path when the narration has to be repeatable or revised later, and it is the path this guide focuses on. Either way, the real work happens afterward in Fairlight, where you set levels, shape the voice with EQ and dynamics, and drop the music under the dialogue.

Most searches stop at "how do I add the audio," but the import is the easy part. Resolve accepts virtually any professional audio format, so a generated file drops straight in. What separates a finished voiceover from a raw file sitting under the picture is the cleanup on the Fairlight page, and that is where this walkthrough spends most of its length. For a presentation, course module, or training video, the training video narration guide covers the writing side before you ever open Resolve.

Two ways to add a voiceover in DaVinci Resolve

Pick the method before you touch the timeline, because the cleanup differs for each one.

Approach	How it works	Best for	Watch out for
Record live in Fairlight	Click a track's input button, choose the microphone, patch it, arm the track with the R button, then record. The take lands on the selected track.	You are the narrator, the script is short, and the room is quiet.	A stumble means another take. Mic placement and room noise affect every recording.
Import generated narration	Create the audio on your Mac, export a common format, drag the file into the media pool, and place it on an audio track.	Tutorials, presentations, and explainers that get revised, or narration you want without a microphone.	You set sync and levels yourself, and you should keep a clean master file outside the project.

If you are the narrator and the take is short, recording live is free and gives you a human performance. If the narration describes steps that may change, a generated file is easier to fix, because you edit one line, regenerate one clip, and swap it in rather than re-recording the whole thing.

Import a generated narration file into the timeline

Write the script before you generate anything, and split it into one short block per scene. Resolve works in clips, and a scene is the smallest unit you will want to redo. Then follow this sequence.

Create the narration and export a common audio format. Resolve works with virtually all professional audio formats, so a WAV or M4A file imported from your Mac drops in without conversion.
Drag the file from the Finder into the media pool, which is the bin at the top left of the screen. You can also drag whole folders and keep their structure.
Switch to the Edit page and drag the clip down onto a dedicated audio track, ideally a track you name Narration so it stays separate from music and effects.
Name each exported file to match its scene, for example 02-export-settings.wav, so it sorts in order in the Finder and the media pool.
If you generated one long take, cut it apart at scene boundaries with the blade tool and nudge each piece until the words land on the right frame.
Save the clean master outside the Resolve project so you can regenerate or re-import a single line without touching your edits.

Keeping one script file and one audio file per scene is what makes the loop reusable. When a fact changes, edit one line, regenerate one clip, and replace it. Murmur is a local-first macOS voice studio for Apple Silicon and Intel Macs running macOS 15 or later, and it exports the narration file you drop into this step. Because it runs the models on your own machine, an unreleased feature name or client detail in the script never leaves the Mac. The audio export guide explains master versus delivery files if you want the fuller picture.

Clean up the narration in the Fairlight page

Click the Fairlight page, then select the narration clip. Every clip in the timeline has its own volume, pan, pitch, and a 6 band parametric EQ, and the timeline combines clip and waveform editing in one view. The order below matters: fix level and noise first, then shape tone, then place the voice against the music.

Level, noise, and sibilance

Set a target level. Right click the clip and choose the normalization command to set a consistent target level across all your narration clips, then fine tune with the clip gain line in the middle of the clip.
Remove steady noise. Drag the noise reduction plug-in from the effects library onto the clip or onto its mixer channel, then either let it detect noise in a selection or select a quiet section and click Learn so it removes what matches.
Cut power line hum with the de-hummer plug-in if your recording picked it up, though clean generated audio usually will not need it.
Tame harsh "s" sounds with the de-esser, which the Fairlight page includes for exactly this purpose.
Watch the loudness meters at the right of the monitoring panel. The meters are color coded blue when levels meet the selected standard, yellow for tolerance, and red when they exceed it.

Generated narration is usually clean, so you may only need the level and the de-esser. Spend the noise tools on real recordings, not on a file that never had noise to begin with.

EQ and dynamics: shape the voice

Double click the EQ area above the channel strip in the mixer to open the 6 band parametric equalizer, and double click the dynamics area to open the dynamics window. From there you have three tools that do most of the work for a spoken voice.

A gentle high pass filter in the EQ to remove low rumble below the voice.
A compressor to even out loud and quiet passages so the narration holds a steady presence.
A limiter to catch any peak that would otherwise clip, set so it rarely engages.
The expander or gate to push down any low level hiss between phrases when you are working with a live recording.

You can adjust the controls, A/B the result, save a preset, and copy the settings to another track, which is handy when every scene should sound the same. For a clean generated file, a light compressor and a true peak safety limiter are often enough. Do this pass after the edit is locked, because moving clips can change what overlaps and what you hear.

Duck the music under the narration

A music bed that sounds fine on its own will bury the voice on laptop speakers. Ducking fixes this automatically. Open the ducker on the music track, which is a track FX plug-in, and choose the narration track as its dialogue source from the dropdown. The ducker then lowers the music whenever there is dialogue and raises it when the voice stops. Adjust the duck level to set how far the music drops, and set the recovery so the music returns at a natural pace. This is the single most useful setting for a narrated presentation, and it is faster and more consistent than drawing volume by hand.

Fit the narration to the picture

A generated take will rarely match your cut exactly. Two Fairlight tools solve that. Elastic wave retiming lets you right click a clip, choose elastic wave, and drag its edge to stretch the length without changing the pitch, so you can line the voice up with the picture instead of re-recording it. For aligning sound to a specific frame, turn on the audio and video scrollers from the timeline view options menu: you get a scrolling filmstrip next to the waveforms, and you can double click a frame to move the playhead and then slide the audio into place.

Here is a short two-scene narration script to import and practice on. Keep one block per scene.

Scene 1, Dashboard tour
Welcome back. This is the dashboard, where every saved project appears as a card. Hover a card to preview it, and click once to open it.

Scene 2, Export settings
Before you export, open Settings. Choose your format and resolution here. The export runs in the background, so you can keep editing while it finishes.

When the narration is longer than the shot, resist speeding up the voice until it sounds rushed. Trim a sentence or give the visual more time. If you mixed on studio headphones, watch the whole project once with sound on the speakers your audience will use, and check the joins between scenes, where an imported clip can start a beat early or end with a tail of silence. If the same narration also needs to go out on video with another voice track, the AI dubbing workflow covers that delivery path.

Where a local voice studio fits

Murmur handles the generation half of this workflow on a Mac. It covers text to speech, voice cloning, voice design, and multi-speaker projects, and it exports the file you import in the steps above. It runs its models locally, so there are no per-character fees and your script is not uploaded, which matters when a presentation mentions an unreleased feature or a client name. If you want the script to audio handoff in one place, the Mac narration workflow walks through it. Murmur costs $49 one-time with a free plan and a 3-day Pro trial, and a 7-day refund policy, and the model list lives in the app, so check it there for the voices and languages you need.

When Murmur is not the right choice

Murmur is a Mac app, so if you edit Resolve on Windows or Linux you will want a different source for the narration. It is also not built for live or real-time streaming speech, and it does not do speech to text, so it cannot transcribe an interview into a script. If you need a specific celebrity voice, or an emotional performance only a person can give, plan on recording a human narrator in Fairlight. And if your team needs several editors working in one shared cloud project, a cloud service fits better. Murmur is strongest when you want private, repeatable narration you can regenerate and drop into a Resolve project you already own.

Frequently asked questions

How do I add a voiceover to DaVinci Resolve without a microphone?

Generate the narration on your Mac as an audio file, export a common format such as WAV or M4A, drag the file into the media pool, and place it on a dedicated audio track in the timeline. A microphone is only needed for the live record option on the Fairlight page.

Where is the Fairlight page in DaVinci Resolve?

Fairlight is one of the pages along the bottom of the window, next to Edit, Cut, Color, Fusion, and Deliver. Click the Fairlight label to switch to it. The Fairlight page is a full audio post environment with a mixer, EQ, dynamics, repair plug-ins, and loudness monitoring.

How do I make the music quieter under my voice?

Use the ducker. Open the ducker track FX on the music track and choose your narration track as the dialogue source. The ducker then lowers the music automatically whenever the voice plays and raises it when the voice stops. Set the duck level for how far the music drops and the recovery for how quickly it returns.

My imported narration is not the right length. Can I stretch it?

Yes. Right click the clip, choose elastic wave, and drag the edge to stretch the clip without changing its pitch. For aligning to a specific frame, turn on the audio and video scrollers from the timeline view options, then slide the audio until it lines up with the filmstrip.

What level should I normalize narration to in Resolve?

Set a target level with the normalization command, which you reach by right clicking the clip, and keep that target consistent across all narration clips. Then watch the loudness meters on the Fairlight page and aim for the blue range, which marks levels that meet your selected standard. Pick a standard that matches where the video will be published.

Sources

Try Murmur - $49 one-time. No subscriptions, no cloud, no per-character fees.

Originally published at murmurtts.com