Voice messages and files to text
Turn a Telegram voice message, a video note, an audio or video file into text. Send it to @AuxioBot and get the transcript back in the same chat.
Send a recording to @AuxioBot and the bot writes out what is said. Nothing needs to be chosen first: the bot starts as soon as the file arrives.
What you can send
- Voice messages and round video notes, sent to the bot or forwarded from any chat.
- Audio files: sent as audio or as a document. The bot recognises MP3, M4A, AAC, OGG, OGA, OPUS, WAV and FLAC.
- Video files: sent as video or as a document. MP4, MOV, MKV, WebM, AVI and M4V are accepted. The bot takes the soundtrack and transcribes it.
Any other document gets the answer Send audio, video, a voice message or a link: this file can't be transcribed. For a link instead of a file, see video links to text.
What comes back
- A short recording (up to about 3,500 characters of text) comes back as a plain message you can copy or forward. No file, no timecodes.
- A longer recording comes back as one message with a collapsible text, the file in your format (TXT, Markdown, Word or PDF), a summary and timecodes. See formats and summary.
Language
The language is detected automatically. For a noisy, very short or mixed-language recording, open Settings (send /settings, or press Transcribe a file on the main menu and then Settings) and turn on Language hint. From then on the bot asks for the language before each file you send in a private chat (not for links, not in groups) and offers Russian, English, Spanish, German, French, Italian, Portuguese, Ukrainian or auto-detect. The hint makes the result noticeably more accurate.
Length and limits
A recording can be up to 6 hours, for everyone. It costs 1 credit per started minute after your free minutes (the first 30 minutes or 5 recordings); if the credits do not cover the whole file, the bot transcribes the part that fits and tells you. The bot transcribes one recording per user at a time: if you send a second one earlier, the bot asks you to wait for the result. See credits and prices.
Who is speaking
For longer recordings, from 10 minutes, the text can be split by speakers with timecodes, when the speech recognition server supports it. In a private chat the bot first asks how many people speak: 1, 2, more than 2 or not sure. The answer makes the speaker labels more accurate; if you are not sure, the bot detects it itself. Shorter recordings are almost always one voice, so they come as plain text.
Questions
Why did the bot say no speech was recognised?
The file has no speech in it, or the sound is silent or too quiet. Check that the recording really contains a voice. A recording with no speech is not charged: the bot returns the credits and free minutes spent on it.
Can I transcribe a recording in a group?
Yes, voice messages and video notes are transcribed in groups automatically. See transcription in groups.
How accurate is the text?
It depends on the sound: a clear voice gives a clean text, noise, music and overlapping speakers lower the quality. The language hint helps when the language is detected wrongly. Names and numbers are worth a check against the audio.