VAENAR-TTS

This repo contains code accompanying the paper "VAENAR-TTS: Variational Auto-Encoder based Non-AutoRegressive Text-to-Speech Synthesis".

Samples | Paper | Pretrained Models

Usage

0. Dataset

1. English: LJSpeech
2. Mandarin: DataBaker(标贝)

1. Environment setup

conda env create -f environment.yml
conda activate vaenartts-env

2. Data pre-processing

For English using LJSpeech:

CUDA_VISIBLE_DEVICES= python preprocess.py --dataset ljspeech --data_dir /path/to/extracted/LJSpeech-1.1 --save_dir ./ljspeech

For Mandarin using Databaker(标贝):

CUDA_VISIBLE_DEVICES= python preprocess.py --dataset databaker --data_dir /path/to/extracted/biaobei --save_dir ./databaker

3. Training

For English using LJSpeech:

CUDA_VISIBLE_DEVICES=0 TF_FORCE_GPU_ALLOW_GROWTH=true python train.py --dataset ljspeech --log_dir ./lj-log_dir --test_dir ./lj-test_dir --data_dir ./ljspeech/tfrecords/ --model_dir ./lj-model_dir

For Mandarin using Databaker(标贝):

CUDA_VISIBLE_DEVICES=0 TF_FORCE_GPU_ALLOW_GROWTH=true python train.py --dataset databaker --log_dir ./db-log_dir --test_dir ./db-test_dir --data_dir ./databaker/tfrecords/ --model_dir ./db-model_dir

4. Inference (synthesize speech for the whole test set)

For English using LJSpeech:

CUDA_VISIBLE_DEVICES=0 TF_FORCE_GPU_ALLOW_GROWTH=true python inference.py --dataset ljspeech --test_dir ./lj-test-2000 --data_dir ./ljspeech/tfrecords/ --batch_size 16 --write_wavs true --draw_alignments true --ckpt_path ./lj-model_dir/ckpt-2000

For Mandarin using Databaker(标贝):

CUDA_VISIBLE_DEVICES=0 TF_FORCE_GPU_ALLOW_GROWTH=true python inference.py --dataset databaker --test_dir ./db-test-2000 --data_dir ./databaker/tfrecords/ --batch_size 16 --write_wavs true --draw_alignments true --ckpt_path ./db-model_dir/ckpt-2000

Reference

1. XuezheMax/flowseq
2. keithito/tacotron