• This is a PyTorch implementation of Microsoft's FastSpeech 2: Fast and High-Quality End-to-End Text to Speech.

• Now supporting about 900 speakers in 🔥 LibriTTS for multi-speaker text-to-speech.

This project supports 2 muti-speaker datasets:

• LJSpeech

• LibriTTS

• VCTK

Configurations are in:

• config/dataset.yaml
• config/hparams.py

Please modify the dataest and mfa_path in hparams.

In this repo, we're using MFA v1. Migrating to MFA v2 is a TODO item.

1. preprocess.py
2. train.py
3. synthesize.py

[DATASET] / wavs / speaker / wav_files [DATASET] / txts / speaker / txt_files

• wav_dir : the folder containing speaker dirs ( [DATASET] / wavs )
• txt_dir : the folder containing speaker dirs ( [DATASET] / txts )
• save_dir : the output directory (e.g. "./processed" )
• --prepare_mfa : create mfa_data
• --mfa : create textgrid files
• --create_dataset : generate mel, phone, f0 ....., metadata.json

• LJSpeech:

#run the script for organizing LJSpeech first
python ./script/organizeLJ.py

python preprocess.py /storage/tts2021/LJSpeech-organized/wavs /storage/tts2021/LJSpeech-organized/txts ./processed/LJSpeech --prepare_mfa --mfa --create_dataset

• LibriTTS:

python preprocess.py /storage/tts2021//LibriTTS/train-clean-360 /storage/tts2021//LibriTTS/train-clean-360 ./processed/LibriTTS --prepare_mfa --mfa --create_dataset

• VCTK:

python preprocess.py /storage/tts2021/VCTK-Corpus/wav48/ /storage/tts2021/VCTK-Corpus/txt ./processed/VCTK --prepare_mfa --mfa --create_dataset

1. spker table
2. traning data
3. validation data

• data_dir : the preprocessed data directory
• --comment: some comments

• LJSpeech:

python train.py ./processed/LJSpeech --comment "Hello LJSpeech" 

• LibriTTS:

python train.py ./processed/LibriTTS --comment "Hello LibriTTS" 

• VCTK:

python train.py ./processed/VCTK --comment "Hello VCTK"

• --ckpt_path: the checkpoint path
• --output_dir: the directory to put the synthesized audios

python synthesize.py --ckpt_path ./records/LJSpeech_2021-11-22-22:42/ckpt/checkpoint_125000.pth.tar --output_dir ./output

• FastSpeech 2: Fast and High-Quality End-to-End Text to Speech, Y. Ren, et al.
• FastSpeech: Fast, Robust and Controllable Text to Speech, Y. Ren, et al.
• xcmyz's FastSpeech implementation
• rishikksh20's FastSpeech2 implementation
• TensorSpeech's FastSpeech2 implementation
• NVIDIA's WaveGlow implementation
• seungwonpark's MelGAN implementation