search

KoljaB / RealtimeSTT

A robust, efficient, low-latency speech-to-text library with advanced voice activity detection, wake word activation and instant transcription.
MIT License
2.09k stars 190 forks source link

Changing to Openai/whisper-large-v3-turbo #135

Open TouficKashmar opened 1 month ago

TouficKashmar commented 1 month ago

What is the process if I want to change into the whisper-large-v3-turbo instead of using faster whisper? Could you please guide me on that?

KoljaB commented 1 month ago

You can type "deepdml/faster-whisper-large-v3-turbo-ct2" as model parameter in RealtimeSTT to try out the new distil model directly. Since RealtimeSTT is based on faster_whisper there is no way of using the lib without it.

homelab-00 commented 1 month ago

As I understand it, any CTranslate2 converted model should -at least in theory- work. As indicated by KoljaB here.

The project uses the faster_whisper library for speech to text, and so it can use only models that are compatible with that library - those being CTranslate2 converted models like faster-whisper or distil-whisper or distil-faster-whisper.

Edit: deepdml/faster-whisper-large-v3-turbo-ct2 does indeed work fine on my system.