A Text-to-Speech Transformer in TensorFlow 2.

Implementation of an autoregressive Transformer based neural network for Text-to-Speech (TTS).

This repo is based on the following paper:

Spectrograms produced with LJSpeech and standard data configuration from this repo are compatible with WaveRNN.


Can be found here.

These samples' spectrograms are converted using the pre-trained WaveRNN vocoder.

The TTS weights used for these samples can be found here.

Check out the notebooks folder for predictions with TransformerTTS and WaveRNN or just try out our Colab notebook:

Open In Colab


Make sure you have:

  • Python >= 3.6

Install espeak as phonemizer backend (for macOS use brew):

sudo apt-get install espeak

Then install the rest with pip:

pip install -r requirements.txt

Read the individual scripts for more command line arguments.


You can directly use LJSpeech to create the training dataset.


  • If training LJSpeech, or if unsure, simply use config/standard
  • EDIT PATHS: in data_config.yaml edit the paths to point at your dataset and log folders

Custom dataset

Prepare a dataset in the following format:

|- dataset_folder/
|   |- metadata.csv
|   |- wav/
|       |- file1.wav
|       |- ...

where metadata.csv has the following format:

Create training dataset

python --config config/standard


python --config config/standard

Training & Model configuration

  • Training and model settings can be configured in model_config.yaml

Resume or restart training

  • To resume training simply use the same configuration files AND --session_name flag, if any
  • To restart training, delete the weights and/or the logs from the logs folder with the training flag --reset_dir (both) or --reset_logs, --reset_weights

Monitor training

We log some information that can be visualized with TensorBoard:

tensorboard --logdir /logs/directory/


from utils.config_manager import ConfigManager
from import reconstruct_waveform

config_loader = ConfigManager('config/standard')
model = config_loader.load_model()
out = model.predict('Please, say something.')

# Convert spectrogram to wav (with griffin lim)
wav = reconstruct_waveform(out['mel'].numpy().T, config=config_loader.config)