search

tigros / Whisperer

Batch speech to text using OpenAI's whisper.
264 stars 24 forks source link

Whisperer 3.0 - "Skip if output exists" maybe not working #37

Closed RickArcher108 closed 11 months ago

RickArcher108 commented 11 months ago

Hi Tigros,

I have a folder in which I had completed transcribing about 1,292 of about 1,800 mp3 files using Whisperer 2.9. This morning when I saw v3.0, I cancelled the 2.9 run and started processing the folder with v3.0. But it didn't recognize that 1,292 had been done. It started from scratch. I couldn't tell whether it was overwriting existing vtt files or what. So, I cancelled that job and resumed it using 2.9, and it recognized that 1,292 had been completed and started processing the rest.

Maybe this feature would have worked if all those files had originally been processed with v3.0. I'll experiment with that once this batch has been completed using 2.9.

Thanks

tigros commented 11 months ago

Hey Rick,

Did you add vtt or something, if any 1 of the checked types don't exist it will redo. On my end seems to work fine.

RickArcher108 commented 11 months ago

My settings look like this:

image

That's in v2.9 but they looked the same in v3.0. When I'm finished with this batch, I'll run a batch with v3.0 and see if the problem recurs.

RickArcher108 commented 11 months ago

I'm afraid v.3 still isn't working for me. It's producing VTT files with nothing in them but the letters WEBVTT. I'm using the ggml-large-v3.bin.

tigros commented 11 months ago

it's using same version of whisper as 2.9, so keep using model v2 not v3.

RickArcher108 commented 11 months ago

Will do. Thanks.