ibab / tensorflow-wavenet

A TensorFlow implementation of DeepMind's WaveNet paper
MIT License
5.42k stars 1.29k forks source link

New Wavenet Paper Has Been Released #306

Open mrsylerpowers opened 7 years ago

mrsylerpowers commented 7 years ago

https://deepmind.com/blog/high-fidelity-speech-synthesis-wavenet/

296

The new, improved WaveNet model still generates a raw waveform but at speeds 1,000 times faster than the original model, meaning it requires just 50 milliseconds to create one second of speech. In fact, the model is not just quicker, but also higher-fidelity, capable of creating waveforms with 24,000 samples a second. We have also increased the resolution of each sample from 8 bits to 16 bits, the same resolution used in compact discs.

It is also now capable of running at scale and is the first product to launch on Google’s latest TPU cloud infrastructure.

jyegerlehner commented 7 years ago

Huzzah, our own ibab is an author on the paper!

ljuvela commented 7 years ago

Awesome, they made a GAN with a WaveNet discriminator! How long until a WGAN version comes out?

eliphatfs commented 7 years ago

Wow, good news!

justking100 commented 7 years ago

@jyegerlehner The real question is if the implementation will be added to this repository.

jyegerlehner commented 7 years ago

@justking100 I didn't mean to suggest they would. Deepmind hasn't open sourced their research results AFAIK. Only some tools like Sonnet. I wouldn't expect them to begin now.

toannhu commented 7 years ago

Hope they will update the implementation of the new Wavenet method to this repo soon :+1:

erdemapaydin commented 7 years ago

i would love an implementation on here!

veqtor commented 6 years ago

Hope we will see the distilled wavenet as a part of the magenta repo

nakosung commented 6 years ago

Was @ibab working at google? 🥇

weixsong commented 6 years ago

Is the IAF network the same as VAE? As mentioned in the paper, 4 transformation of the logistic distribution (input) transfer the input to teacher distribution, is each smaller student WaveNet need to predict mean and var to do information?