FastSpeech 2

Implementation of "FastSpeech 2: Fast and High-Quality End-to-End Text to Speech"

Quick Start

1. Prepare dataset

   mkdir -p data/raw/
   cd data/raw/
   wget https://data.keithito.com/data/speech/LJSpeech-1.1.tar.bz2
   tar -zxf LJSpeech-1.1.tar.bz2
   cd ../../
   python datasets/tts/lj/prepare.py

2. Forced alignment

   # Download MFA first: https://montreal-forced-aligner.readthedocs.io/en/stable/aligning.html
   # unzip to montreal-forced-aligner
   ./montreal-forced-aligner/bin/mfa_train_and_align data/raw/LJSpeech-1.1/mfa_input data/raw/LJSpeech-1.1/dict_mfa.txt data/raw/LJSpeech-1.1/mfa_outputs -t ./montreal-forced-aligner/tmp -j 24

3. Build binary data

   # fs2
   PYTHONPATH=. python datasets/tts/lj/gen_fs2.py --config configs/tts/lj/fs2.yaml
   # fs2s
   PYTHONPATH=. python datasets/tts/lj/gen_fs2s.py --config configs/tts/lj/fs2s.yaml

4. Train FastSpeech 2 and 2s

   CUDA_VISIBLE_DEVICES=0 python tasks/fs2.py --config configs/tts/lj/fs2.yaml --exp_name fs2_exp1 --reset
   
   CUDA_VISIBLE_DEVICES=0 python tasks/fs2s.py --config configs/tts/lj/fs2s.yaml --exp_name fs2s_exp1 --reset

5. Download pre-trained vocoder

   mkdir wavegan_pretrained
   

   download checkpoint-1000000steps.pkl, config.yml, stats.h5 from https://drive.google.com/open?id=1XRn3s_wzPF2fdfGshLwuvNHrbgD0hqVS to wavegan_pretrained/

6. Inference

   CUDA_VISIBLE_DEVICES=0 python tasks/fs2.py --config configs/tts/lj/fs2.yaml --exp_name fs2_exp1 --infer
   CUDA_VISIBLE_DEVICES=0 python tasks/fs2s.py --config configs/tts/lj/fs2s.yaml --exp_name fs2s_exp1 --infer