I appreciate your response. I should not have replied in the manner that I did. I think there are many ways to make the argument against using these libraries, and all of the maintainers would agree it doesn't make sense in many situations. I am fine with these arguments, but I don't think it warrants referring to the work we've done as half-baked or hobby projects. Even if not the intention, I think it implies a lack of effort and professionalism which comes across as more disparaging than simply pointing out the gaps.
It has always been possible to do custom transformations via Series.transform. And here is a PR adding it to dataframes: https://github.com/elixir-explorer/explorer/pull/912 - the PR uses existing APIs, it was possible before, it just makes it more convenient. I am sure more will be added later
To be fair to the creators of “half baked” libraries, you’re comparing our work to frameworks with several years head start with the backing of trillion dollar organizations (excluding polars). To have convinced ANYBODY to successfully make the switch is, to me, both somewhat of a miracle and an indication that there is a legitimate market for this work. I think it also dispels the hobby claim
As for your issues with Explorer, perhaps it is not the library, but your knowledge of Elixir that was a little too half baked? ;)
Yeah the latency in my demo is definitely worse. There also seem to be some issues where it picks up it’s own audio and keyboard/mouse clicks and tries to transcribe them which leads to derailed convos.
The unnatural voices happen I think because for some reason the way I stream audio to the browser breaks up the audio from ElevenLabs, so you get almost this “start/stop” sound that makes it sound worse than usual.
There’s a lot of minor details to get right, but it’s a fun problem to try to solve. I wish Retell the best :)
It should not be too hard! From what I saw Whisper is similar to Bart, and we have Bart. The missing piece is a library for audio processing to tensors.
No need to convert the format! Bumblebee supports loading parameters directly from PyTorch. You can just specify `{:local, path/to/checkpoint}` in `Bumblebee.load_model`
It’s a pleasant surprise to see this shared here, I am the author of this piece. Honestly I wrote this post for myself more than anything else. I also find that my knowledge in a lot of areas is very “surface level.” It’s really easy to regurgitate definitions, but it’s definitely harder to get to the core of those ideas. I hope you enjoyed!