Telegram voice to text: transcripts, summaries and tasks from voice notes

There are three ways.

1. Telegram's built-in feature. A transcribe button next to the voice message. Unlimited with Telegram Premium; without it, a few transcriptions a week.
2. A transcription bot. Forward the voice note, get text back. Plenty of free options.
3. An AI assistant. It transcribes and then works with the text right away: summary, tasks, a reply.

If you only occasionally want to read a voice note instead of listening, the first or second option is enough. The third is for when the voice note needs to turn into action.

Without Premium, Telegram gives a limited number of transcriptions per week. When the feature opened to everyone in December 2023, the limit was two a week. Premium has no limit.

Telegram sets the current number in its server configuration and can change it, so check the counter in your own app.

You send a voice note to the chat with your assistant — recorded or forwarded.

• Speech is recognised by a Whisper model (the faster-whisper implementation) running on the Agent OS server. The audio isn't sent to a third-party transcription service.
• The language is detected from the speech itself, not from your Telegram interface language. Ukrainian speech with a Russian interface is still recognised as Ukrainian.
• Limits: up to 8 minutes and 20 MB per message.
• Besides voice notes, the assistant understands audio files, videos and round video messages — both the picture and the speech.

What happens next depends on the caption. Forward a voice note with the caption "transcribe" and you get the text word for word. Without a caption, the assistant carries out what was said as a task by default — if it was a request rather than a thought out loud, say so in the caption.

This is where an assistant differs from a plain transcriber.

• Call notes. The "Call notes" skill lists topics, agreements, and who owes what by when — only what was actually said.
• Tasks and reminders. Say "remind me tomorrow at 10 to call the supplier" and the reminder is set.
• A page from a voice note. The "Page from a voice note" skill builds an HTML page file from an idea you dictated.
• Notes to memory. Dictate price list changes and the assistant remembers them.

If a recording cuts off or is unclear, the assistant says so instead of filling in the gaps.

Yes, in client mode on PRO and above. A customer sends your bot a voice note, the assistant recognises it and answers from the facts in your project. You get a card with the question and the answer.

Client mode has a daily cap: by default, 20 replies to outside people per assistant.

Speech recognition runs on the Agent OS server with a local model. The resulting text is then processed by an OpenAI model to carry out your task; conversations aren't used for training.

The bot is closed to outsiders: only you can message it until you open a group or client mode yourself.

There's no separate voice plan: a voice note is a request like any text message and spends credits from your monthly budget. START is $29/mo with 3,000 credits.

The first 3 days are free: 150 credits and no automatic charge. If you only need a couple of transcriptions a week, Telegram's built-in feature is the honest choice.

Telegram's transcription limits come from the Telegram API documentation on voice transcription and Engadget (01.12.2023); the recognition engine is described in the faster-whisper repository on GitHub. Voice limits, language detection and skills come from the Agent OS code as of 24.09.2026. Prices come from the live checkout catalogue on the same date.

• An AI secretary on Telegram
• Ready-made "Personal assistant" setup: voice notes, summaries, reminders
• How much an AI assistant on Telegram costs: 2026 pricing guide
• AI assistant services on Telegram compared, 2026
• Checklist: how to launch an AI agent on Telegram