News for hackers. Stuff that matters.
Re:Git Diagramming "The Weave" by Xmd5a on Sunday August 31, 2025 @07:19AM (#45082313)
Thank you! I studied linguistics, this is why I'm aware of this reality. I'm asking this because this is HN and chances that your worked on speech-to-texts systems is rather high. I was wondering how this problem is tackled in systems such as Whisper. I haven't used it much, but I definitively noticed that YouTube's transcription model try to strike a balance between transcription fidelity and readability. Sometimes it will elude some pauses & repairs to produce a more streamlined transcription, sometimes it will include them. It's an art I guess, but a model where you could tune this at inference time would be interesting.

There are no comments on this comment yet.