search

tigros / Whisperer

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

Openai whisper small model #66

Closed tjesp closed 4 months ago

tjesp commented 4 months ago

I had the chance to try the original Openai Whisper on my linux system. I cannot use the large model since my graphics card is 8GB.

However I found that generally the small model in Openai whisper were as accurate as the large model in whisperer.

I thought it was supposed to just be a faster implementation than openai whisper but it seems that some things might be lost in the translation.

Do you have any comments on that could it be improved any way?

tigros commented 4 months ago

Hi,

if i recall original Openai Whisper is CPU only, so why you need graphics memory? it's python right? maybe you mean RAM.

ya i use base myself even smaller than small, works fine. maybe large can handle subtle speech a bit better.

tjesp commented 4 months ago

No Openai Whisper is python but it can use graphics card acceleration via pytorch. Thanks I am closing this now.