Tacotron 2 with efficient batching

This repo is built based on NVIDIA's PyTorch implementation of Tacotron2, which speeds up the TTS model training by 40+%, without sacrificing model performance. The algorithm is general to any seq2seq model training with variable length input, and can also be applied to other training tasks such as in ASR and speaker recognition.

For the details of the batching algorithms, please refer to our publication: Speed up training with variable length inputs by efficient batching strategies in Interspeech 2021.

Currently, this branch is identical to the NVIDIA's PyTorch implantation. It serves as the starting point of implementing our proposed batching strategies. This work is a part of our on-going speech synthesis and voice cloning project, and we are actively working on making this part available on top of a "clean" version of Tacotron2. The actual implementation code of the efficient batching algorithm will be released by Sep. 10th, 2021, along with a detailed README for example usages.

Please come back to check out later.

====== The content below is copied over from the original NVIDIA PyTorch Tacotron2 ======

Tacotron 2 (without wavenet)

PyTorch implementation of Natural TTS Synthesis By Conditioning Wavenet On Mel Spectrogram Predictions.

This implementation includes distributed and automatic mixed precision support and uses the LJSpeech dataset.

Distributed and Automatic Mixed Precision support relies on NVIDIA's Apex and AMP.

Visit our website for audio samples using our published Tacotron 2 and WaveGlow models.

Pre-requisites

1. NVIDIA GPU + CUDA cuDNN

Setup

1. Download and extract the LJ Speech dataset
2. Clone this repo: git clone https://github.com/NVIDIA/tacotron2.git
3. CD into this repo: cd tacotron2
4. Initialize submodule: git submodule init; git submodule update
5. Update .wav paths: sed -i -- 's,DUMMY,ljs_dataset_folder/wavs,g' filelists/*.txt
   • Alternatively, set load_mel_from_disk=True in hparams.py and update mel-spectrogram paths
6. Install PyTorch 1.0
7. Install Apex
8. Install python requirements or build docker image
   • Install python requirements: pip install -r requirements.txt

Training

1. python train.py --output_directory=outdir --log_directory=logdir
2. (OPTIONAL) tensorboard --logdir=outdir/logdir

Training using a pre-trained model

Training using a pre-trained model can lead to faster convergence
By default, the dataset dependent text embedding layers are ignored

1. Download our published Tacotron 2 model
2. python train.py --output_directory=outdir --log_directory=logdir -c tacotron2_statedict.pt --warm_start

Multi-GPU (distributed) and Automatic Mixed Precision Training

1. python -m multiproc train.py --output_directory=outdir --log_directory=logdir --hparams=distributed_run=True,fp16_run=True

Inference demo

1. Download our published Tacotron 2 model
2. Download our published WaveGlow model
3. jupyter notebook --ip=127.0.0.1 --port=31337
4. Load inference.ipynb

N.b. When performing Mel-Spectrogram to Audio synthesis, make sure Tacotron 2 and the Mel decoder were trained on the same mel-spectrogram representation.

Related repos

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

nv-wavenet Faster than real time WaveNet.

Acknowledgements

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

We are inspired by Ryuchi Yamamoto's Tacotron PyTorch implementation.

We are thankful to the Tacotron 2 paper authors, specially Jonathan Shen, Yuxuan Wang and Zongheng Yang.