mazoza

I dont actually see any tokens used in the model. It seems like the model actually predicts latents and then VAE converts back to audio. More like Tortoise or XTTS

Can one of the authors explain what this actually means from the post?

hertz-vae: a 1.8 billion parameter transformer decoder which acts as a learned prior for the audio VAE. The model uses a context of 8192 sampled latent representations (17 minutes) and predicts the next encoded audio frame as a mixture of gaussians. 15 bits of quantized information from the next token act as semantic scaffolding to steer the generation in a streamable manner.

meh this is not that good. Sounds quite boring.

I've been using Coqui Studio for a while now to generate AI voices for my video game project. It delivers most that is in the blogpost. I like it but I'd ask for a better API support if anyone sees this from Coqui.

It is a lot faster, has more languages and models and can run inference on even CPU with many of their models.

Coqui is hosting a week-long competition/hackathon for STT in long-tail languages.

You will be able to train STT models for your favorite languages using CoquiSTT

...and you get free GPU (V-100) access in order to run your experiments.

the competition lasts one week and already started!

I means it is faster than real time almost 10x

So it is the contrary

I know the old speech team continues as Coqui https://github.com/coqui-ai/

It is sad seeing another Mozilla project going down to the hole.

Also, why do they pay people to develop on something that is not being maintained?