Keyword Timestamp Identification in Milliseconds

[stevenmills]

The accuracy on Whisper in a few of my tests has been excellent. One drawback I have is the timestamp ranges provided do not infer enough accuracy. For example I've seen 6 word sentences within the audio take ~1 second to speak, but the range given on the phrase is 7 seconds long. I am in need of the start milliseconds and duration milliseconds of each word. Is there a technical limitation to Whisper being able to provide this information per word?