uttertype
uttertype copied to clipboard
[Feature Request] Help remove hallucinations based on "open mic" time
Thanks for a great project!
Is your feature request related to a problem? Please describe. As a user, I get hallucinations when the mic is left open during pauses and at the end of my recordings. Generally weird throw-ins like: "Please see the complete disclaimer at https://sites.google.com" or "For more information, visit www.FEMA.gov"
Describe the solution you'd like I recognize that these take place at the OpenAI level, but it would be great if uttertype were able to help strip these hallucinations somehow.
Additional context
It seems like no_speech_prob could be a potential pathway to explore?
https://github.com/openai/whisper/discussions/928#discussioncomment-7255165