Sanskrit transcription: a rough draft
Sanskrit is one of the weakest languages in the speech model. It heard very little spoken Sanskrit in training, so expect a rough draft with many corrections, and sometimes lines that drift into Hindi or Marathi spelling. Slow, clearly spoken Sanskrit from one speaker, as in a class or a lecture, works best.
Chanted and sung recitation is harder still. Melody, long held vowels and fixed rhythm do not match ordinary speech, so the draft for a stotra, a Gita chapter or a Vedic recitation can be far from the text. For well-known texts, the printed source is the better reference, and you can use the transcript mainly to find timings.
Upload the recording as it is: MP3, M4A, WAV, OGG, OPUS, FLAC or AAC, a voice note saved from WhatsApp or Telegram, or a video file such as MP4 or MOV. You can also record from your microphone in the browser. Long files are split and processed in parallel, so a ten-minute recording usually takes a minute or two.
Devanagari, sandhi and compounds
The transcript uses Devanagari. Points to check while reviewing:
- Sandhi joins words at their boundaries, and the draft may split or join them differently from a printed edition.
- Long compounds may be broken into separate words.
- Visarga (ः), anusvara (ं) and halanta (्) are often missing or misplaced.
- Long and short vowels can be swapped.
Transliteration and other scripts
There is no built-in IAST or other Roman transliteration, and the transcript is not produced in other Indic scripts. Correct the Devanagari first, then convert it with a separate tool if you need IAST.
Lectures about Sanskrit are often given in Hindi, English or a regional language with Sanskrit quotations. In that case select the main language of the talk and fix the quoted verses by hand.
What people record in Sanskrit
Uploads include spoken Sanskrit classes, university and gurukula lectures, discourses, shloka practice sessions, and recordings made by students who want to follow along with a text. Teachers use the draft as a base for class notes.
Exports and next steps
Export TXT or DOCX for notes and documents, PDF for a fixed copy, SRT or VTT for subtitles, and CSV with one row per line and its start time. Timestamps and speaker names can be switched on or off in each export. Speaker labels are available on paid plans or with a Minute Pack.
For video, such as a recorded class or a YouTube discourse, see the Sanskrit video to text page.