Hacker News new | ask | show | jobs
by okeysmokey 546 days 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.
1 comments

Yes. 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.
How are you doing speaker id?
It (mostly correctly) ID'd the SCOTUS justices on this one. Pretty cool! https://transcriberai.com/Overview/aa908e33-5680-462a-94ff-6...
Really cool! thanks for sharing.