[celltalk]

With this logic, you should also be able to trim the parts that doesn’t have words. Just add a cut-off for db, and trim the video before transcription.

Possibly another 10-20% gain?