Saik0s / Whisperboard

The open-source iOS app that's making quality voice transcription more accessible on mobile devices.
GNU General Public License v3.0
755 stars 78 forks source link

Support for large-v2 #10

Closed Mario03482 closed 1 year ago

Mario03482 commented 1 year ago

Will you add the support for large-v2? Maybe on the new iPhones the large model can run without any issue?

Saik0s commented 1 year ago

Yes! It will already in the repository and will be in a new update.

Mario03482 commented 1 year ago

Is there a deadline?

Saik0s commented 1 year ago

Submitted new version to App Store, waiting for review. I expect it to be available today - tomorrow.

Mario03482 commented 1 year ago

Another question: I wanted to use this app to transcribe long audio file(~2h) using the large-v2, do you think it is possible to do it without having any bug/crash? iPhone 14 pro max

Ps: do you have any benchmark with the large model on a latest iPhone?

Saik0s commented 1 year ago

A new version is now available on the App Store!

I haven't tested large files yet, so I don't anticipate any bugs, except for the time it takes to transcribe audio. I believe that for such lengthy files, I need to implement a resuming mechanism so users can continue transcription after exiting the app. I've included live progress updates for the transcription, which should help users understand the current stage. I plan to add some benchmarks later.

Mario03482 commented 1 year ago

What is you to do list now? And the priorities at the moment?

Saik0s commented 1 year ago

I have several things that I want to focus:

coder543 commented 1 year ago

I see a "large" model in the app, but there's no indication of whether it is Large-V1 or Large-V2. I tried glancing through the source code here, and I'm still not sure. Since "V2" doesn't appear anywhere in here, I'm wondering if it might still be V1.

Either way, this seems like a really neat app!

Saik0s commented 1 year ago

It is V2, I decided to not include V1. Will add this to description of the model.