[Feature] Confidence score #75
Description
Feature Description
omniASR returns raw transcripts, but the API makes it difficult to extract the probability associated with the model's confidence in certain speech.
Having outputs enriched with log probabilities/confidence scores (like whisper) would be very beneficial to the community in order to:
- Know if the language of the speech is not supported by the model
- Get an idea of the overall sound quality
- Be able to detect and remove false positives
Use Case
No response