Produce Audio with Convolutional neural networks

Question

Were there any approaches on Producing Audio with Convolutional neural networks?

There are a lot of approaches on producing images through convnets. But i see no articles or posts about producing audio.

According to this topic on stackoverflow , the post writer says :

"I have found out the audio can be represented as spectrograms."

So why cant it be done?

To do this with Convnets, should I :

a) Use LSTM with the conv layers?

B) What should be the output? Considering the spectogram...

Answer 1

Yes, of course there is, WaveNet by Deepmind produces audio that is very similar to "human-like" quality. It can also generate music.

https://deepmind.com/blog/wavenet-generative-model-raw-audio/