You can hear text spoken in Hindi, Spanish, French, and other supported languages on your Android phone without a cloud speech service. OGAM (Off Grid AI Mobile) Pro generates the voice on your device. Download the voice resources first, choose a language and speaker, then play an assistant message while offline.
Get OGAM on Google Play | Mobile features and Pro
Use it to listen to an explanation, review a short answer, or hear a passage in a language you are learning. The procedure below reads assistant messages inside OGAM. It does not replace Android's system-wide screen reader or add a read-aloud button to every app.
What do you need for offline text-to-speech on Android?
You need OGAM Pro, a downloaded local voice model, and the assets for the voice you select. To generate a new AI reply offline, you also need a downloaded local text model. The text model writes the answer; the voice model speaks it.
Use version 0.0.111 or later for these settings. The published Android requirements are Android 10 or later and at least 4 GB of RAM.
Before disconnecting, finish:
- Installing the app and activating Pro.
- Downloading the on-device voice model.
- Selecting the language and voice you plan to use.
- Downloading a local text model if you want new replies while offline.
Switching to another language can require more downloads. A downloaded English voice does not mean every language is ready without internet. Wait for the selected voice's download and loading to finish before testing offline playback.
Which languages and voices can you choose?
The mobile voice runtime provides English with US and UK voices, plus French, Spanish, Italian, Portuguese, Hindi, Polish, and German. The app shows the voices supported by its installed runtime. Select the language first to see the matching speakers.
| Language choice | Voice names |
|---|---|
| English (US) | Heart, River, Sarah, Adam, Michael, Santa |
| English (UK) | Emma, Daniel |
| French | Siwis |
| Spanish | Dora, Alex |
| Italian | Sara, Nicola |
| Portuguese | Dora, Santa |
| Hindi | Alpha, Omega, Psi |
| Polish | Mateusz |
| German | Anna |
This list reflects the mobile runtime used by the current app. It is separate from speech-to-text's 99 languages label. Recognizing a language from audio and generating a voice in that language are different capabilities.
The underlying mobile speech runtime documentation describes its multilingual voice models. The language and speaker choices shown in your app are the ones to use for your installed version.
How do you download and select a local voice?
Open Models > Voice and download the on-device voice model. Its model name is Kokoro TTS. Then open Chat Settings > TEXT TO SPEECH to enable playback and choose the language and speaker. Keep the selection local for offline use.
1. Download the voice model
In Models, choose Voice. Select the on-device model card and complete its download. If a remote voice was active, select the local card to switch back to the phone's voice model.
The download includes the resources needed by the selected voice. Leave enough free storage for those files and any other models you use.
2. Enable speech in Chat mode
Open Chat Settings > TEXT TO SPEECH. In Chat mode, switch on Enable TTS. This enables speech controls for assistant messages.
If the section says no voice models are downloaded, finish the download in Models > Voice first.
3. Choose the language
Under Language, select the language of the text you want spoken. For example, choose Hindi for Hindi text or Spanish for Spanish text.
Wait if the app shows a voice download or switching status. The app may need a language-specific model or pronunciation resources before it can use your selection.
4. Choose a speaker and speed
Select a voice from the list for that language. Start at 1.0x speed. The Speed setting runs from 0.5x to 2.0x, so you can slow a passage for listening practice or increase the pace for familiar material.
Changing speed does not change the language or translate the words.
How do you hear a reply while completely offline?
After setup, turn on airplane mode and check that Wi-Fi is off. Open a chat with a downloaded local text model and ask for a short reply in the selected voice's language. When the answer finishes, long-press the assistant message and select Speak.
For a first check, use a request such as:
Reply in Spanish. Explain how to make a cup of tea in four short sentences.
Select a Spanish voice before playing that answer. Use a language you can check, or compare the written answer with a trusted reference.
You can also use the speech control on the assistant message when it is available. The Speak action appears for completed assistant messages when text-to-speech is enabled and the voice is ready.
You do not need microphone input for this process. Typing a request and hearing the answer uses text generation and text-to-speech. Recording your own voice would add a separate speech-to-text step.
Can you read your own text aloud?
The message action described here reads assistant replies. If you have a short passage of your own, you can paste it into a local chat and ask the model to return it unchanged. Compare the returned text with your original before selecting Speak.
For example:
Return the passage below exactly as written. Do not translate, summarize, or add an introduction.
A chat model can still change text despite that instruction. Check names, numbers, and wording when exact reproduction matters. This is a chat-based workflow, not a dedicated document narration or audio export tool.
Does choosing a Spanish voice translate English text?
No. The voice setting chooses how supplied text is spoken. It does not translate the text. To hear a passage in another language, first use a suitable local text model to translate it, review that translation, then choose a voice for the translated language.
The same rule applies to an AI answer. If you want a French answer spoken in French, ask the text model to answer in French and select a French voice.
A voice model's language support does not establish the text model's translation quality. Check the words before relying on the spoken result.
Why does a selected voice fail or sound wrong?
Check that its download finished and its language matches the text. Then check the active voice provider and playback speed. An English voice selected for a Hindi passage is a different issue from a missing file or an unavailable remote service.
| Problem | What to check |
|---|---|
| Speak is missing | Enable TTS in Chat Settings, wait for the voice to become ready, and use a completed assistant message. |
| A new language needs internet | Its voice or pronunciation resources may not be downloaded yet. Complete setup before disconnecting. |
| Speech uses the wrong pronunciation | Match Language and Voice to the actual text. |
| Playback is too quick or slow | Return Speed to 1.0x, then adjust it. |
| No sound is audible | Check Android media volume and the selected speaker or connected headphones. |
| Speech works online but not offline | Check that the selected voice model is on-device, rather than a remote voice service. |
| No new answer appears while offline | Select a downloaded local text model as well as the local voice. |
Test a short paragraph before a long answer. Names, abbreviations, and sentences that mix languages can need closer review. Selecting one voice does not promise natural pronunciation for every language inside the same passage.
Is multilingual text-to-speech free?
Text-to-speech is part of OGAM Pro. The free mobile app includes speech-to-text, which turns your voice into words. Spoken replies are the opposite direction and use the Pro voice feature.
Once Pro is activated and the selected voice resources are downloaded, local speech generation does not need a cloud speech API. A remote voice provider or remote text model changes that network requirement.
Choose one language, finish its download, and play one short reply with the phone offline. Then prepare any other languages you want before you need them away from a connection.
Top comments (0)