search

kaixxx / noScribe

Cutting edge AI technology for automated audio transcription. A nice GUI for OpenAIs Whisper and pyannote (speaker identification)
GNU General Public License v3.0
440 stars 80 forks source link

Update NoScribe Function #97

Open Lod3 opened 3 hours ago

Lod3 commented 3 hours ago

This might be early but a way to let users know about an update might significantly improve transcription accuracy or speed considering the leaps between WhisperV2 and Whisper3. It might be a good idea to prevent future discussions about this to let users decide what Whisper version they want to use, or what language, but I think that was referenced in another issue.

Some notification in the app would be nice to know if there is a new version first. A way to download the new version and install it over the current one would improve the update process. Currently I remove noscribe and then download the new version to update. I can see a lot of users do not want to go this process and keep using an inferior version that would improve the workflow significantly.

If there is a way to detect if the user is using a dedicated GPU with enought VRAM and point/guide them to the relevant nsisbin file to improve transcription file to add that would be usefull.

gernophil commented 3 hours ago

Thanks for the suggestions. I also have one, @kaixxx: maybe we could significantly reduce the download size by including download scripts for the models instead of the model itself. Then users could decide wihich one to download. We could simply store those models in the same folder as the config.yml then. What do you think? Regarding the updates: this will be for windows only since for Mac it's common practice to always fully replace the app. Don't think any other way is possible, but for windows this might be an option.

Lod3 commented 2 hours ago

@gernophil good suggestion, like a small setup file and then the scripts could run during the setup wizard. https://github.com/bugbakery/audapolis does something similar

image If a user can check off the models they want here they cannot use the software untill the download is done though. And how do you add models or change them afterwards?

kaixxx commented 29 minutes ago

a way to let users know about an update

This is already implemented with version 0.5: If I release a new version on GitHub, users will be notified on startup. However, the feature has not yet been used since I have not made an official new release since then.

A way to download the new version and install it over the current one would improve the update process.

You can do this already. The installer (on Windows at least) will suggest uninstalling the old version and do this automatically if you choose this option. However, you can also choose to install directly over the old version.

If there is a way to detect if the user is using a dedicated GPU with enought VRAM and point/guide them to the relevant nsisbin file to improve transcription file to add that would be usefull.

The autodetection of GPUs is tricky and may cause errors. So I prefer that users actively decide to use CUDA, at least for now. Catching possible incompatibilities and suggesting solutions could be improved, though.

@gernophil

including download scripts for the models instead of the model itself. Then users could decide wihich one to download.

I have to think about this. Given that most of the users don't know much about the pros and cons of different models available, such a decision would probably be too much for them in a situation where they just want to get noScribe up and running and test it out. So I prefer to offer a complete package that just works out of the box. But it would be great to include an easy way (in the UI) to download and install other models later on. This would probably also mean curating a list of models that are known to work in noScribe.