Closed evanxqs closed 1 month ago
There are hallucinations exist when using of Triton whisper with TensorRT-LLM in sherpa,and a VAD can solve most of hallucinations as my experience, so i wonder is there VAD in Triton whisper that i can enable it directly? or i have to integrate a new one by myself?
You may integrate it a new one by checking this https://github.com/k2-fsa/sherpa-onnx/blob/master/python-api-examples/vad-with-non-streaming-asr.py. Welcome to contribute if you do so.
Or you could do VAD at the client side.
There are hallucinations exist when using of Triton whisper with TensorRT-LLM in sherpa,and a VAD can solve most of hallucinations as my experience, so i wonder is there VAD in Triton whisper that i can enable it directly? or i have to integrate a new one by myself?
You may integrate it a new one by checking this https://github.com/k2-fsa/sherpa-onnx/blob/master/python-api-examples/vad-with-non-streaming-asr.py. Welcome to contribute if you do so.
silero-vad https://github.com/snakers4/silero-vad
thank you ! and do you know which vad is better?
There are hallucinations exist when using of Triton whisper with TensorRT-LLM in sherpa,and a VAD can solve most of hallucinations as my experience, so i wonder is there VAD in Triton whisper that i can enable it directly? or i have to integrate a new one by myself?
You may integrate it a new one by checking this https://github.com/k2-fsa/sherpa-onnx/blob/master/python-api-examples/vad-with-non-streaming-asr.py. Welcome to contribute if you do so.
silero-vad https://github.com/snakers4/silero-vad
thank you ! and do you know which vad is better?
They are exactly same model. It's up to you.
There are hallucinations exist when using of Triton whisper with TensorRT-LLM in sherpa,and a VAD can solve most of hallucinations as my experience, so i wonder is there VAD in Triton whisper that i can enable it directly? or i have to integrate a new one by myself?