alesaccoia / VoiceStreamAI

Near-Realtime audio transcription using self-hosted Whisper and WebSocket in Python/JS
MIT License
736 stars 106 forks source link

OpenAI Whisper model is not working #22

Closed jinmiaoluo closed 4 months ago

jinmiaoluo commented 5 months ago

When I specify the type as whisper instead of fast_whisper using the --asr-type, an error is reported in the browser's console. The error message is as follows:

image

image

The OpenAI Whisper model does not support the words attribute. Consequently, the value of transcript_data.words is a string of UNSUPPORTED_BY_HUGGINGFACE_WHISPER instead of null. We should either check the type of transcript_data.words or set the words field in transcript_data to null.