Flowtron: an Autoregressive Flow-based Network for Text-to-Mel-spectrogram Synthesis

Rafael Valle, Kevin Shih, Ryan Prenger and Bryan Catanzaro

In our recent paper we propose Flowtron: an autoregressive flow-based generative network for text-to-speech synthesis with control over speech variation and style transfer. Flowtron borrows insights from Autoregressive Flows and revamps Tacotron in order to provide high-quality and expressive mel-spectrogram synthesis. Flowtron is optimized by maximizing the likelihood of the training data, which makes training simple and stable. Flowtron learns an invertible mapping of data to a latent space that can be manipulated to control many aspects of speech synthesis (pitch, tone, speech rate, cadence, accent).

Our mean opinion scores (MOS) show that Flowtron matches state-of-the-art TTS models in terms of speech quality. In addition, we provide results on control of speech variation, interpolation between samples and style transfer between speakers seen and unseen during training.

Visit our website for audio samples.

Pre-requisites

1. NVIDIA GPU + CUDA cuDNN

Setup

1. Clone this repo: git clone https://github.com/NVIDIA/flowtron.git
2. CD into this repo: cd flowtron
3. Initialize submodule: git submodule update --init; cd tacotron2; git submodule update --init
4. Install PyTorch
5. Install python requirements or build docker image
   • Install python requirements: pip install -r requirements.txt

Training from scratch

1. Update the filelists inside the filelists folder to point to your data
2. Train using the attention prior and the alignment loss (CTC loss) until attention looks good python train.py -c config.json -p train_config.output_directory=outdir data_config.use_attn_prior=1
3. Resume training without the attention prior once the alignments have stabilized python train.py -c config.json -p train_config.output_directory=outdir data_config.use_attn_prior=0 train_config.checkpoint_path=model_niters
4. (OPTIONAL) If the gate layer is overfitting once done training, train just the gate layer from scratch python train.py -c config.json -p train_config.output_directory=outdir train_config.checkpoint_path=model_niters data_config.use_attn_prior=0 train_config.ignore_layers='["flows.1.ar_step.gate_layer.linear_layer.weight","flows.1.ar_step.gate_layer.linear_layer.bias"]' train_config.finetune_layers='["flows.1.ar_step.gate_layer.linear_layer.weight","flows.1.ar_step.gate_layer.linear_layer.bias"]'
5. (OPTIONAL) tensorboard --logdir=outdir/logdir

Training using a pre-trained model

Training using a pre-trained model can lead to faster convergence. Dataset dependent layers can be ignored

1. Download our published Flowtron LJS, Flowtron LibriTTS or Flowtron LibriTTS2K model
2. python train.py -c config.json -p train_config.ignore_layers=["speaker_embedding.weight"] train_config.checkpoint_path="models/flowtron_ljs.pt"

Fine-tuning for few-shot speech synthesis

1. Download our published Flowtron LibriTTS2K model
2. python train.py -c config.json -p train_config.finetune_layers=["speaker_embedding.weight"] train_config.checkpoint_path="models/flowtron_libritts2k.pt"

Multi-GPU (distributed) and Automatic Mixed Precision Training (AMP)

1. python -m torch.distributed.launch --use_env --nproc_per_node=NUM_GPUS_YOU_HAVE train.py -c config.json -p train_config.output_directory=outdir train_config.fp16=true

Inference demo

Disable the attention prior and run inference:

1. python inference.py -c config.json -f models/flowtron_ljs.pt -w models/waveglow_256channels_v4.pt -t "It is well know that deep generative models have a rich latent space!" -i 0

Related repos

WaveGlow Faster than real time Flow-based Generative Network for Speech Synthesis

Acknowledgements

This implementation uses code from the following repos: Keith Ito, Prem Seetharaman and Liyuan Liu as described in our code.