great idea, we'll do that too. we just decided to launch an onnx first and get some feedback. we'll be simplifying the process of running it everywhere including a command line executable.

yes, indeed. we are working on adding mit licensed phonemizers too by this weekend, so you'll be able to use these models as you like :)

I think you misunderstood the question. I guess its only open-weights not open-source then.

For some insight into the original question, take a look at the Debian ML policy:

https://salsa.debian.org/deeplearning-team/ml-policy

hey sorry for this issue, i think its a bug in our preprocessing. let me look into it and help fix it. i think you posted this in our discord so lets carry the conversation there.

for that, a 100KB model could be enough ;)

glad you liked it, thank you so much for the kind words. our team is really good at squeezing performance out of small models. we are working on a new launch and hope to release a technical report along with that which includes details. fyi, our current 14M model is better than our previous 80M model. and we expect this trend to continue.

i can confirm that we did not.

What's the source of that voice then in training data? It sounds insanely close to her voice. Strange parallel to openAI denying they trained on Scarlet J's voice.

thanks a lot for the feedback. glad you liked it. we're gonna be launching more tiny models across use-cases.

thanks for pointing these errors out. we're looking into this and will help fix this.

yeah let me add uv and conda support to make it easier.

Thanks! I asked my bot to make me a plugin for it and it one-shotted it, the resulting script was ~20 lines, very nice!

One of the most responsive developers I’ve ever seen, kudos

damnn, really sorry for the inconv, looks like some folks are having bad env issues. we're working on fixing this.

It's absolutely not your fault. It's a skill issue and compatibility issue on my end and/or python. You guys are doing amazing.