Critical issues with VAC

[SilasK]

1. Once an end is detected the buffer up to the end of speack frame is added to the online audio buffer, but never transcribed.

https://github.com/QuentinFuxa/whisper_streaming_web/blob/e3550ef07dc819e93773068de963d5c02316aa89/whisper_streaming_custom/online_asr.py#L316

2. If a start is detected the online ASR is re-initialized. but the previous audio buffer might not be finished!

https://github.com/QuentinFuxa/whisper_streaming_web/blob/e3550ef07dc819e93773068de963d5c02316aa89/whisper_streaming_custom/online_asr.py#L396C29-L396C33

I found a solution to this by creating new entities of online-asr for each speak segment and finish one after the other. https://github.com/SilasK/realtime-translation/blob/0f2e7f368548eea3370d3acf5d44d8f236be23df/src/whisper_streaming/online_asr.py#L554

[SilasK]

I saw that in the main server you have a nice logic with many task. Should VAC be a individual task that runs before the transcription and the diarizaiton?

[lifengmds]

any updates ?

[QuentinFuxa]

Hey can you try with https://github.com/QuentinFuxa/WhisperLiveKit/releases/tag/0.1.7 ? VAC is functional in this released

[lifengmds]

thanks, i tried with your the latest updates , the VAC is working now.