Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

The option parameter does not affect the VAD process. #916

Closed
zh-plus opened this issue Jul 19, 2024 · 2 comments · Fixed by #923
Closed

The option parameter does not affect the VAD process. #916

zh-plus opened this issue Jul 19, 2024 · 2 comments · Fixed by #923

Comments

@zh-plus
Copy link
Contributor

zh-plus commented Jul 19, 2024

After #856, the VAD-related option paramter does not affect the vad results at all.

options: Optional[NamedTuple] = None,

I believe self.options isn't used during VAD. Should we modify VadOptions, and make it take effect here?

def load_vad_model(self, vad_onset=0.500, vad_offset=0.363):

@Jiltseb
Copy link
Contributor

Jiltseb commented Jul 22, 2024

I think you are confused between VadOptions (which is a silero VAD related parameter) and options variable (that is used to initialize the inference pipeline with default transcription options).

Batched version of faster whisper does not use silero VAD at the moment, hence the VadOptions variable is not needed for VAD.
If required, it is possible to make the vad_onset and vad_offset user configurable for tuning the VAD model in batched inference.

@zh-plus
Copy link
Contributor Author

zh-plus commented Jul 22, 2024

Thank you for your response! I just noticed the difference.

I think it would be beneficial to make vad_onset and vad_offset user-configurable. This flexibility would allow users to adjust these parameters according to their specific data requirements.

I may take a PR for this issue soon.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging a pull request may close this issue.

2 participants