seanmor5

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.

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? ;)

Hey, I'm the author! All of the chapters are done, but there are still some minor updates as APIs change. It should be going to production soon

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 :)

You can use Torchx with the `mps` device, but many ops are not implemented for mps

For some of it! I was thinking more something for loading different audio encodings to Nx tensors…so doing everything up until Nx

There is a project in the ecosystem that builds on top of MLIR and can target any MLIR dialect, but it is not mature enough yet.

I think it would be great to see!

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`

We are all really into bumblebees, specifically the kind that can transform into 1977 Yellow Camaros