sotelo/parrot

RNN-based generative models for speech.

Speech synthesis using recurrent neural networks.

This repo has the code for our ICLR submission:

Jose Sotelo, Soroush Mehri, Kundan Kumar, João Felipe Santos, Kyle Kastner, Aaron Courville, Yoshua Bengio. Char2Wav: End-to-End Speech Synthesis.

The website is here.

NOTE: The code is currently being refactored/cleaned/documented. We wanted to make it available as soon as possible but we know well that the current version is not ready for replication. Therefore, if you're interested in this, please come back later. Alternatively, you can send me an email and I will let you know when it's ready.

NOTE(2): The code for the neural vocoder is based on sampleRNN.

Updates:

• SampleRNN modules added
• End-to-end model can generate more than 4s of audio in 1s on p6000 gpu. (This is achieved by generating 200 samples of length 10s in a single batch. Total time taken for this is 448 seconds.)

parrot's Issues

fuel datasets

The referenced fuel datasets ['arctic', 'blizzard', 'dimex', 'librispeech', 'pavoque', 'vctk'] are not in the fuel distribution. Are there standard converters for any of these already in other projects?

Why not provide the simple tutorial in README?

As a green hand in Python and Theano, I want to know the command line to run the demo training the network according to the data set such as VCTK and generating the audio files. It is just a job of several line.

Example hyperparameters

Though you haven't yet done an extensive hyperparameter search, might you give us one or two examples of hyperparameter settings you have used?

How to build/run parrot?

Cool paper!

But in as far as this project/code goes (in github) I see lots of dependancies, such as 'blocks', and no guide how to run/repro the claimed results.

Any chance the authors can provide a bit more info?

-best

Understand the cost

Hi,

Just wondering if you can explain more about the cost value for the model.

From line, seems the cost of the READER is ignored, or I am missing something here.

Thanks,

Jian

website link at the readme should be http, cause https doesn't work

http://www.josesotelo.com/speechsynthesis/
not
https://www.josesotelo.com/speechsynthesis/

Is pretrained model is available?

Is pretrained model is available?

When will the model be ready to the public?

Hi,

This project is really promising and I am eagerly waiting for the completion of it. While what do you mean by saying ' the current version is not ready for replication'? Does it mean some codes are missing or the structure of the code is kind of messy?

Best,