DEV Community: XTSoft

I stopped missing half my standups — notes on live bilingual captions without a meeting bot

XTSoft — Wed, 23 Sep 2026 07:57:23 +0000

I work in a company where the meeting language is not my native one.

Standups are fine on a good day. Design reviews are worse: three people talking over each other, acronyms I do not know yet, and someone saying “ship it” while I am still parsing the previous sentence. Native meeting captions help sometimes. They also fail in the ways that matter when you are already behind:

They are often host-gated — if the organizer never turns them on, you get nothing.
They are usually one language at a time — I want the original and a translation side by side when a term is ambiguous.
They do not help me take a note without leaving the conversation — by the time I open a doc, the thread has moved on.
The “AI notetaker” path often means a bot joins the call — extra participant, recording notice, IT policy questions, and a little social tax every time.

So I stopped optimizing for “better post-call transcript” and started optimizing for staying present while the call is still happening.

The constraint that changed the product shape

Offline meeting tools optimize for a clean recording and a polished summary afterward.

Live tools optimize for a different set:

Work on Google Meet, Zoom, and Microsoft Teams without waiting for each platform’s feature flag.
Latency low enough that captions feel attached to the voice — for me, roughly a second is the line between aid and distraction.
Original + translation available together when I want learning mode, not only a single target language.
Notes tied to the same timeline as the captions, so “that sentence about Redis” is clickable later.
Prefer a participant-side capture over inviting another identity into the room.

That last one is the fork in the road. Bot-join tools are easy to demo. They are also the reason some teams never install them.

Why "no bot" is not just a privacy slogan

When a bot joins, three things happen at once:

Everyone sees an extra participant (or a recording banner).
Policy and consent become a meeting-level decision, not a personal tool choice.
You inherit every platform’s bot permissions model.

Participant-side captions flip that: you install something that listens on your machine or in your browser tab. Other people do not get a new invitee. You still need to be honest about what is transcribed for you, but you avoid the social and IT blast radius of a room-wide bot.

That choice has engineering consequences. Browser tab capture covers Meet and web Zoom/Teams well. The Zoom or Teams desktop client does not live in a tab — so a Windows desktop listener became the twin path for “every meeting on this PC,” same account, same minutes pool.

Instant vs accurate translation is a real UX split

Early on I treated “translation quality” as one knob.

In a live meeting it is two jobs:

Instant: keep latency ultra-low for back-and-forth. You accept that a half-finished clause might look rough.
Accurate: wait for a fuller sentence, spend a bit more model time, get better technical terms and phrasing.

I switch between them in the same call. Standup banter wants Instant. A design review full of service names wants Accurate. Shipping only one mode means half your meetings feel wrong.

Captions that become notes beat a second app

The pause-loop for meetings is not scrubbing a video. It is:

hear something important → panic → open Notion → lose the next two sentences → never find the quote later.

Tapping a caption line into a note, with the transcript kept after the call, closed that loop better than “remember to write it down.” Asking the meeting a question afterward (“what did we decide about caching?”) only works if the live capture was good enough to build a timeline on.

Summaries are useful. They are not a substitute for being able to read the line that was said at 00:17.

What I would tell myself before building

Solve missing the live conversation, not “AI meeting notes” as a category.
Treat bot vs participant-side as a product decision, not an implementation detail.
Ship bilingual as a first-class mode — original-only and translation-only are modes, not the whole product.
Accept that browser + desktop are two listeners for one account if you care about Zoom/Teams native apps.
Auto-stop when the room goes quiet. Forgetting a tab open should not silently burn your monthly minutes.

The thing I actually use

I packaged this workflow as SonicMeet — Chrome/Edge extension, optional Windows desktop app, freemium meeting hours.

I am not doing a pricing pitch here. The interesting part for me was treating live meeting understanding as a latency + bilingual layout + no-bot problem, not only an ASR accuracy race.

Open questions

When do you keep Instant mode on by default vs nudge people toward Accurate for technical calls?
How do you explain desktop audio capture without sounding like spyware?
For bilingual captions on a laptop: original on top, translation on top, or a user preference you change mid-call?

If you have built realtime speech tools for meetings — or you have your own ritual for surviving standups in a second language — I would love to hear what you landed on in the comments.

Built this because I needed it. Feedback: support@sonicmeet.app

I stopped reading subtitles and started listening — notes on real-time AI dubbing in a browser tab

XTSoft — Sat, 12 Sep 2026 11:59:44 +0000

For a long time I treated foreign-language video like a reading assignment.

Open the tab. Turn on captions if they exist. Pause when the line is too fast. Scrub back. Open a dictionary. Miss the next joke anyway.

That works for a recorded lecture you control. It falls apart for a livestream, a Coursera week you are already behind on, or a meeting where people keep talking while you are still parsing the last sentence.

Subtitles help you decode. They do not always help you stay present. At some point I wanted the opposite posture: keep my eyes on the video, and let my ears take the translation.

So I started building toward real-time AI dubbing on whatever audio is already playing in the browser — not “upload a file and wait for an MP4.”

The job is not “translation quality” in the abstract

Offline tools optimize for a clean transcript and a polished voiceover. Live tools optimize for a different constraint set:

Stay on the same tab — no export, no special player
Latency low enough to feel watchable — for me, roughly half a second is the line between tool and distraction
Original + translation available together when I want them (learning mode)
A spoken translation when I am tired of reading
Fail gracefully when the network blips — reconnect without throwing away the session mental model

If you only chase BLEU scores, you can still ship something people rage-quit in minute three of a Twitch stream.

Tab audio is a blunt but useful boundary

Site-by-site integrations never end. YouTube today, Coursera tomorrow, some random course player next week.

For v1, “if the tab is playing sound” is a blunt MVP line that actually ships. Capture the tab’s audio, chunk it sensibly (silence is not worth uploading), show captions, and — when dubbing is on — play a translated voice while optionally ducking the original.

That last part matters more than I expected. A dubbed voice at 100% with the original also at 100% is mush. Ducking the source while the AI voice speaks is a small control with a large “this feels intentional” effect.

Dual subtitles changed how I learn; dubbing changed how I watch

Bilingual captions (source + target) cut how often I alt-tab to a dictionary. I still use them when I want accuracy or vocab.

Dubbing is for a different night: long stream, tired eyes, I mostly want to hear the meaning and glance at the picture. Same pipeline, different output surface.

I also learned that “subtitle style” is not cosmetic. A cinema burn-in and a side panel transcript are different reading postures. One keeps you in the video; the other is closer to notes.

What I would tell myself before building

1. Solve the pause loop, not “AI dubbing” as a vibe.

The job-to-be-done was staying present on live or long-form speech.

2. Universal tab audio beats perfect site hacks for v1.

Special-case forever and you never ship Twitch.

3. Voice choice and mix are product, not polish.

People abandon tools that sound wrong even when the text is fine.

4. Desktop is a different problem.

Browser tab capture covers most courses and streams. Downloaded files, desktop Zoom, and “everything on the PC” need a different listener. I ended up with a Windows 11 app for that case — same account, same hours — because the extension alone cannot hear what never enters a tab.

The thing I actually use

I packaged this workflow as DubTab — Chrome/Edge extension, optional Windows app, freemium minutes.

Site: dubtab.com

Chrome Web Store: DubTab

Short demo: YouTube

I am not doing a pricing pitch here. The interesting part for me was treating live dubbing as a latency + mix + session problem, not only an ASR problem.

Open questions

Where do you split live captions vs archival transcripts in product and storage?
How do you explain tab-audio permissions without sounding like malware?
For learners: default to bilingual captions, dubbing, or both — what feels least noisy on a 13" laptop?

If you have built realtime speech tools — or you have your own pause-rewind ritual for foreign streams — I would love to hear what you landed on in the comments.

Built this because I needed it. Feedback: support@dubtab.com

I kept pausing Japanese streams — so I built tab-audio captions on Cloudflare Workers

XTSoft — Wed, 09 Sep 2026 01:49:53 +0000

I'm a Chinese speaker living in the US. Weekdays are mostly English: standup on Zoom, design docs, Slack threads that somehow always explode at 5pm. Nights and weekends are Japanese — VTuber streams, cooking livestreams, the occasional anime episode I refuse to watch with English-only subs.

For months my "language practice" looked like this:

Twitch / YouTube open in Chrome
Host says something fast
I miss it
Pause
Scrub back 8–12 seconds
Still not sure → open a dictionary tab
Stream moves on without me

That loop is survivable for a recorded course. It is miserable for a live stream where chat is already three jokes ahead.

I didn't want another upload-a-video translator. I wanted: stay on the same tab, see the original line + a Chinese (or English) translation almost immediately, keep watching.

So I built a Chrome/Edge extension that captions whatever audio is playing in the tab, and put the realtime path on Cloudflare Workers + Durable Objects.

The product constraint (before the stack)

Hard requirements from day one:

Works on any tab with audio — YouTube, Twitch, Netflix-in-browser, Zoom/Meet in the browser, random course sites
No file upload
Bilingual overlay (source + target)
Latency low enough that a live stream still feels watchable (~0.5s ballpark for me)
Prefer keeping raw audio in the tab; only ship text / small chunks for processing

That ruled out "export MP4 → wait → download SRT." The extension had to own capture; the edge had to own the session.

How the pieces fit (CF-shaped)

High level, nothing fancy:

Extension (Chrome MV3)

Uses tabCapture / offscreen document patterns to grab the tab's audio stream, run VAD-ish chunking so I'm not uploading silence, and render a floating bilingual overlay. Settings and history live locally first.

Worker (stateless edge)

Auth, plan checks, routing. A request comes in: "here's a short audio/text segment for session X." The Worker doesn't hold long-lived state — it just verifies the user and hands the work to the right place.

Durable Object (one per caption session)

This is the part that made the architecture click for me.

Each Start click maps to a Durable Object keyed by session id. That object:

Owns the WebSocket to the extension (hibernation-friendly)
Queues / serializes ASR + translation steps so chunks don't arrive out of order on screen
Tracks minute usage for the free tier without racing a shared Redis
Keeps a short rolling buffer of recent caption lines (enough to recover a reconnect without rebuilding history from scratch)
Tears down cleanly when the user hits Stop or the tab dies

I tried the "just shove everything through a single Worker + external WebSocket server" version first. It worked until two tabs and a flaky Wi‑Fi night. Per-session Durable Objects made the failure domain obvious: one stuck session doesn't take down everyone else's captions.

Why Cloudflare specifically

I didn't want to babysit a caption server in us-west-2 while I'm watching a stream at 1am PT. Workers put the handshake close to me; Durable Objects give me single-threaded session semantics without standing up Redis + a sticky-session LB for a side project that started as "please stop pausing."

ASR / translation providers sit behind the Worker as normal HTTPS calls. The interesting bit for me wasn't swapping models — it was keeping the session glue boring and correct at the edge.

What "good enough" latency felt like

I didn't optimize for leaderboard numbers. I optimized for: can I leave this on for a 90-minute cooking stream without rage-pausing?

For me, captions landing around half a second after speech is the line between "tool" and "distraction." Faster is nicer; much slower and my thumb goes back to the spacebar.

So the goal stayed: watchable live captions, not court-ready transcripts.

A normal evening with it

Typical flow now:

Install the extension
Open the Japanese stream (or the English Zoom tab for work)
Pick target language — Chinese when I'm tired, English when I'm taking notes
Hit Start
Read along: original on top, translation under it

Meetings: browser Meet/Zoom tab → Start → stop pretending I caught every acronym.

Language nights: dual subtitles on, dictionary tab closed unless I really want a deep dive later.

I still export / copy history when I need searchable notes after a lecture. Live comprehension and archival transcripts turned out to be different jobs — which is why the Durable Object keeps only a short buffer, not a forever archive in SQLite by default.

Privacy, said plainly

"Listens to tab audio" sounds creepy if you care about privacy (I do — this runs on my work laptop too).

Design intent:

Capture stays in the extension / tab path
The session DO sees segments for that session, not a permanent dump of every stream I've ever watched
UI makes it obvious when captioning is running

Not a formal security audit. Just: I wouldn't ship something I refuse to enable myself.

The thing I actually shipped

All of that is packaged as SonicCaption — Chrome/Edge extension, freemium minutes, bilingual overlay.

Site: soniccaption.com

Short demo: YouTube

I'm not doing a pricing pitch here. The interesting part was the constraint set and putting session state in Durable Objects instead of "yet another Node process."

What I'd tell myself six months ago

1. Fix the pause loop, not "AI translation" as a vibe.

The job was staying present on a live Japanese stream while my brain is half in English from work Slack.

2. One Durable Object per session beats a clever global queue.

Ordering, reconnects, and usage counters got simpler when the unit of isolation matched the unit of user intent (Start → Stop).

3. Universal tab audio beats site-specific hacks for v1.

Special-casing YouTube forever is how you never ship Twitch.

4. Dual subtitles change behavior.

Source + target together cut how often I alt-tabbed to a dictionary mid-stream.

Open questions

Where do you split live captions vs archival transcripts in product and in storage?
How do you explain tabCapture permissions without sounding like malware?
For bilingual UI on a 13" laptop: stacked lines, side-by-side, or toggle — what feels least noisy?

If you've built realtime stuff on Workers / Durable Objects (or fought the same pause-rewind ritual on JP streams), I'd love to hear what you landed on in the comments.

Built this because I needed it. Feedback: support@soniccaption.com