3 ms·
Is this using Whisper or something else? It looks like the site performs multiple steps on the audio and did a decent job of guessing who was speaking.
by okeysmokey 2y ago
Is this using Whisper or something else? It looks like the site performs multiple steps on the audio and did a decent job of guessing who was speaking.
- WinH 2y agoYes. We run run Whisper Large V3 (not Turbo) for the speech to text. It still seems to be the best open source model out there for that step. The main challenge we are trying to solve is Speaker Identification, which is a very time consuming process.
- okeysmokey 2y agoHow are you doing speaker id?
- okeysmokey 2y agoIt (mostly correctly) ID'd the SCOTUS justices on this one. Pretty cool! https://transcriberai.com/Overview/aa908e33-5680-462a-94ff-6ded511bf613 https://transcriberai.com/Overview/aa908e33-5680-462a-94ff-6...
- LunaRoot 2y agoReally cool! thanks for sharing.