3 ms·
For this use case, why not use Whisper to transcribe the audio, and then an LLM to do a second step (summarization or answering questions or whatever)? If you
by rahimnathwani 11mo ago
For this use case, why not use Whisper to transcribe the audio, and then an LLM to do a second step (summarization or answering questions or whatever)?
If you need diarization, you can use something like https://github.com/m-bain/whisperX https://github.com/m-bain/whisperX
- pants2 11mo agoWhisper simply isn't very good compared to LLM audio transcription like gpt-4o-transcribe. If Gemini 3 is even better it's a game-changer.
- crazysim 11mo agoSince Gemini seems to be sucking at timestamps, perhaps Whisper can be used to help ground that as an additional input alongside the audio.