Silences are not detected, Whisper thinks speech is non-stop

[manumaan]

I want to start by saying big thanks to making it possible to make subtitles in very short time on mac m1 using GPU.

But there is one problem that bugs me.
Say the climax of a mystery movie, where the detective is going to declare the murderer.

Actual script is like:

1:30:00 - 1.30.15 "And so the murderer is.."
Silence of 10 second while the camera sweeps over the suspects
1:30:25 - 1:30:28 "The Butler!"

But when transcribed by Whisper, the silence is not taken into account. The second dialogue comes just as the 1st one ends.

1:30:00 - 1.30.15 "And so the murderer is.."
1:30:16 - 1:30:28 "The Butler!" (Camera is sweeping over suspects for 10 seconds of this time, with no speech in movie)

The words are very accurately (most of the time) detected, but the whisper does not like silences. It starts next dialogues just as soon as the last dialogue ends. This requires some manual effort to fix on the subtitles. Is there any way to fix this?

[thewh1teagle]

This PR should fix it #2279

[manumaan]

That would be great!