Whisper transcription.

[TheStalke]

Hello, Im trying to transcribe a video, for subtitles using Whisper by OpenAi, but at the start of the video theres a song and my .srt file starts from "00:00:00,00" , when theres no one talking like it goes from 0 to 13 but people only start talking from 8 to 13, is there a way to make the subtitles start at the same time as they start talking in the video automatically?

[jonathgh]

Hi @TheStalke, we managed to do this with post-processing, because the timestamps that Whisper produces aren't that accurate. We arrived at a solution that works pretty well- we use VAD (voice activity detection) to check when a speaker is speaking and trim the segments automatically to match:

Before:

After:

Here's the link if you want to try it out: Automatic Segment Trimming

[TheStalke]

Thank you 🌹I will try it out

[misutoneko]

Yes, VAD usually helps. Other than that, you could try with --word_timestamps True.
There are other parameters that can be changed (logprob, compression_ratio, etc.) but their usage isn't super straightforward.

[TheStalke]

Thank you 🌹I will try it out

[TheStalke]

It works thank you 🌹