News for hackers. Stuff that matters.
Re:I built a tool to track what AI leaders say and how their... by iafine on Tuesday October 13, 2026 @11:52PM (#50082188)
YouTube's auto-generated captions are a great product, but one problem is that they can't tell speakers apart in podcasts with multiple people. I'm building something that also works with YouTube transcripts, and it has the same problem.
Re:I built a tool to track what AI leaders say and how their... by gcsydney on Tuesday October 13, 2026 @11:53PM (#50082193)
Thanks for the feedback.
Re:I built a tool to track what AI leaders say and how their... by aavisangle on Wednesday October 14, 2026 @12:29AM (#50082390)
You could use some smaller models, like parakeet, to figure out the users.
Re:I built a tool to track what AI leaders say and how their... by shermantanktop on Wednesday October 14, 2026 @12:55AM (#50082553)
It’s called “diarization“ and the leading models are not nearly as accurate as the leading speech to text - whisperx etc.
Re:I built a tool to track what AI leaders say and how their... by jakderrida on Wednesday October 14, 2026 @05:10AM (#50084623)
If a video is less than 45 minutes, you can plug it into any gemini and have it transcribe and diarize the video. (although, I've been told it only works on aistudio anymore) Sometimes it trips up. However, plugging in videos longer than 45 minutes quickly leads to ludicrous behavior the longer it is over 45 minutes.

You can even just plug in the Youtube link with a short prompt.

Bear in mind that thinking that plugging in the transcripts will help the Gemini model in any way frequently results in mistranscriptions being included and actually compromises the diarizations.

The prompt I use is...

"Please transcribe, diarize, and speaker identify the speakers from the Youtube video."

Another thing I've learned is that long-worded prompts also interfere with the results in weird ways.