Voice
Turning a voice message or a round video into text, offline and without Telegram Premium. The action is called Convert voice to text.
Voice to text
- Voice language — the language of the speech. Only one is used at a time, so this row appears once a second model is installed; with exactly one installed, that one is used.
- The installed models are listed under it. Activating one offers to delete it.
- Download a voice model — the last row. Twenty-three languages are offered, Persian and English first.
No model ships inside the app, because even the small ones are around 50 MB. The first time you convert a voice message, the download for the chosen language is offered with its size, and the same message is then converted.
What to expect
- A different language is never substituted for the one you asked for. Speech through the wrong model does not fail — it returns fluent nonsense — so the app refuses instead.
- Expect the sense of what was said rather than every word exactly. These are the small offline models; casual speech at phone quality is hard for them.
- For an exact reading of an important message, use Summarize voice on the AI screen instead: that one sends the audio to a recognition service and is far more accurate, but it costs a request and sends the recording off the device.