Diarization · overlap separation · ASR · translation · 27 languages

Who said what, and when.

Upload a long recording with several speakers. Every speaker turn is found, overlapping voices are pulled apart, and each turn is transcribed in its speaker's language — and, if you like, translated.

What happens to your recording

Who spoke when. Anuvani-Diarization finds every speaker turn, up to four voices at once, any number of speakers.
Which language. Two independent language-ID models vote on every stretch of speech; each speaker gets their own language.
Overlaps pulled apart. Where voices overlap, a separation network gives each speaker their own audio, guided by the diarization, so nobody's words land in someone else's turn.
Every turn transcribed. INDIC-TRANSCRIBE writes each turn in code-mixed, romanized or native script, checked for truncation and repetition.
Translated, if you ask. Indic Translate reads the transcript in passages of about ten minutes of speech, so every turn is translated with what was said around it, into as many languages as you choose.

You get an interactive transcript in sync with the audio, its translations side by side, and a bundle with your audio, the transcript (JSON, TXT, CSV), subtitles (SRT, WebVTT) in every language and the diarization (RTTM). Correct any turn in place; open the bundle here later to see it again without processing.

Processing

…

0.0s elapsed

Nothing is uploaded

Open earlier results

Drop the bundle you downloaded (.trankit.zip), or the audio together with a transcript.json, an .rttm, or .srt / .vtt subtitles. Everything is read in your browser.

Drop the files here bundle .zip · or audio + transcript.json / .rttm / .srt / .vtt

History

Recordings processed on this server. Results are kept for 7 days; download the bundle to keep one for good.

Recording	Processed	Length	Speakers	Language	Status