Awesome

Tacotron over MXNet

A tech demo of MXNet capabilities consisting of a Tacotron implementation. This is a work in progress.

This project was made during the 8 weeks from 10-2017 to 12-2017 at the PiCampus AI School in Rome.

List of functionalities and TODOs

• Multithreading data iterator
• DSP tools
• CBHG module for spectrograms
• Basic seq2seq example for string reverse. It we'll be used as Tacotron backbone
• Encoder with CBHG
• Attention model
• Custom decoder for processing r * mel_bands spectrograms frames for each time step during the cell unrolling
• Switch to MXNet 1.0
• Switch to Gluon
• Clean up and organize code for better understanding

Getting Started

• install MXNet: <code> pip install -r requirements.txt </code>
• run: <code> python tacotron.py </code>

Using the default setting, a simple dataset will be used as training. Predictions samples will be generated at the end of the training phase.

If you want to train over a big dataset, Kyubyong has cut and formatted <a href="http://www.audiotreasure.com/webindex.htm"> this </a> English bible. You can find his dataset <a href="https://www.dropbox.com/s/nde56czgda8q77e/WEB.zip?dl=0"> here </a> and the CSV text <a href="https://www.dropbox.com/s/lcfhs1kk9shvypj/text.csv?dl=0">here</a> .

Prerequisites

This project has been developed on

• MXNet 0.12
• librosa

Authors

This project was developed by Alberto Massidda and Stefano Artuso during Pi School's AI programme in Fall 2017.

Acknowledgments

• Thanks to Roberto Barra Chicote for supporting us
• Thanks to Keith Ito https://github.com/keithito, Kyubyong Park https://github.com/Kyubyong for making us start diving in