A Text-to-Speech Transformer in TensorFlow 2

Implementation of a non-autoregressive Transformer based neural network for Text-to-Speech (TTS).
This repo is based, among others, on the following papers:

• Neural Speech Synthesis with Transformer Network
• FastSpeech: Fast, Robust and Controllable Text to Speech
• FastSpeech 2: Fast and High-Quality End-to-End Text to Speech
• FastPitch: Parallel Text-to-speech with Pitch Prediction

Our pre-trained LJSpeech model is compatible with the pre-trained vocoders:

• MelGAN
• HiFiGAN

(older versions are available also for WaveRNN)

For quick inference with these vocoders, checkout the Vocoding branch

Non-Autoregressive

Being non-autoregressive, this Transformer model is:

• Robust: No repeats and failed attention modes for challenging sentences.
• Fast: With no autoregression, predictions take a fraction of the time.
• Controllable: It is possible to control the speed and pitch of the generated utterance.

🔈 Samples

Can be found here.

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

Try it out on Colab:

Updates

• 06/20: Added normalisation and pre-trained models compatible with the faster MelGAN vocoder.
• 11/20: Added pitch prediction. Autoregressive model is now specialized as an Aligner and Forward is now the only TTS model. Changed models architectures. Discontinued WaveRNN support. Improved duration extraction with Dijkstra algorithm.
• 03/20: Vocoding branch.

📖 Contents

• Installation
• API
• Dataset
• Training
  • Aligner
  • TTS
• Prediction
• Model Weights

Installation

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.

Pre-Trained LJSpeech API

Use our pre-trained model (with Griffin-Lim) from command line with

python predict_tts.py -t "Please, say something."

Or in a python script

from data.audio import Audio
from model.factory import tts_ljspeech

model = tts_ljspeech()
audio = Audio.from_config(model.config)
out = model.predict('Please, say something.')

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

You can specify the model step with the --step flag (CL) or step parameter (script).
Steps from 60000 to 100000 are available at a frequency of 5K steps (60000, 65000, ..., 95000, 100000).

IMPORTANT: make sure to checkout the correct repository version to use the API.
Currently 493be6345341af0df3ae829de79c2793c9afd0ec

Dataset

You can directly use LJSpeech to create the training dataset.

Configuration

• If training on LJSpeech, or if unsure, simply use config/training_config.yaml to create MelGAN or HiFiGAN compatible models
  • swap the content of data_config_wavernn.yaml in config/training_config.yaml to create models compatible with WaveRNN
• EDIT PATHS: in config/training_config.yaml edit the paths to point at your dataset and log folders

Custom dataset

Prepare a folder containing your metadata and wav files, for instance

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

if metadata.csv has the following format wav_file_name|transcription you can use the ljspeech preprocessor in data/metadata_readers.py, otherwise add your own under the same file.

Make sure that:

• the metadata reader function name is the same as data_name field in training_config.yaml.
• the metadata file (can be anything) is specified under metadata_path in training_config.yaml

Training

Change the --config argument based on the configuration of your choice.

Train Aligner Model

Create training dataset

python create_training_data.py --config config/training_config.yaml

This will populate the training data directory (default transformer_tts_data.ljspeech).

Training

python train_aligner.py --config config/training_config.yaml

Train TTS Model

Compute alignment dataset

First use the aligner model to create the durations dataset

python extract_durations.py --config config/training_config.yaml

this will add the durations.<session name> as well as the char-wise pitch folders to the training data directory.

Training

python train_tts.py --config config/training_config.yaml

Training & Model configuration

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

Resume or restart training

• To resume training simply use the same configuration files
• 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

tensorboard --logdir /logs/directory/

Prediction

With model weights

From command line with

python predict_tts.py -t "Please, say something." -p /path/to/weights/

Or in a python script

from model.models import ForwardTransformer
from data.audio import Audio
model = ForwardTransformer.load_model('/path/to/weights/')
audio = Audio.from_config(model.config)
out = model.predict('Please, say something.')

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

Model Weights

Access the pre-trained models with the API call.

Old weights

Model URL	Commit	Vocoder Commit
ljspeech_tts_model	0cd7d33	aca5990
ljspeech_melgan_forward_model	1c1cb03	aca5990
ljspeech_melgan_autoregressive_model_v2	1c1cb03	aca5990
ljspeech_wavernn_forward_model	1c1cb03	3595219
ljspeech_wavernn_autoregressive_model_v2	1c1cb03	3595219
ljspeech_wavernn_forward_model	d9ccee6	3595219
ljspeech_wavernn_autoregressive_model_v2	d9ccee6	3595219
ljspeech_wavernn_autoregressive_model_v1	2f3a1b5	3595219

Maintainers

• Francesco Cardinale, github: cfrancesco

Special thanks

MelGAN and WaveRNN: data normalization and samples' vocoders are from these repos.

Erogol and the Mozilla TTS team for the lively exchange on the topic.

Copyright

See LICENSE for details.