A pytorch implementation of Tacotron: A Fully End-to-End Text-To-Speech Synthesis Model.

• Install python 3
• Install pytorch == 0.2.0
• Install requirements:

  pip install -r requirements.txt
  

I used LJSpeech dataset which consists of pairs of text script and wav files. The complete dataset (13,100 pairs) can be downloaded here. I referred https://github.com/keithito/tacotron for the preprocessing code.

• hyperparams.py includes all hyper parameters that are needed.
• data.py loads training data and preprocess text to index and wav files to spectrogram. Preprocessing codes for text is in text/ directory.
• module.py contains all methods, including CBHG, highway, prenet, and so on.
• network.py contains networks including encoder, decoder and post-processing network.
• train.py is for training.
• synthesis.py is for generating TTS sample.

• STEP 1. Download and extract LJSpeech data at any directory you want.
• STEP 2. Adjust hyperparameters in hyperparams.py, especially 'data_path' which is a directory that you extract files, and the others if necessary.
• STEP 3. Run train.py.

• STEP 1. Run synthesis.py. Make sure the restore step.

• You can check the generated samples in 'samples/' directory. Training step was only 60K, so the performance is not good yet.

• Keith ito: https://github.com/keithito/tacotron

• Any comments for the codes are always welcome.