AdaSpeech - PyTorch Implementation

This is an unofficial PyTorch implementation of AdaSpeech. AdaSpeech: Adaptive text to speech for custom voice.

This project is based on ming024's implementation of FastSpeech 2.

Note:

• Support multi languague training, the default phoneme support Vietnamese and English, custom for other language
• Utterance level encoder and Phoneme level encoder to improve acoustic generalization

• Conditional layer norm which is the soul of AdaSpeech paper

Requirements:

• Install Pytorch Before installing pytorch please check your Cuda version by running following command : nvcc --version

pip install -r requirements.txt

Training

Preprocessing

• First, align the corpus by using MFA tool to get TextGrid (note that you have to run each language separately then move all speaker's TextGrid in to single folder named "textgrid")
• copy textgrid folder in to preprocessed path

run the preprocessing script

python preprocess.py config/pretrain/preprocess.yaml

Training

Train baseline model with

python train.py [-h] [-p PREPROCESS_CONFIG_PATH] [-m MODEL_CONFIG_PATH] [-t TRAIN_CONFIG_PATH] [--vocoder_checkpoint VOCODER_CHECKPOINT_PATH] [--vocoder_config VOCODER_CONFIG_PATH]

Finetune

Preprocessing

First, align the corpus by using MFA tool to get TextGrid (note that only finetune 1 speaker for best quality)

run the preprocessing script

python preprocess.py config/finetune/preprocess.yaml

Finetune

Finetune speaker voice with

python finetune.py [-h] [--pretrain_dir BASE_LINE_MODEL_PATH] [-p PREPROCESS_CONFIG_PATH] [-m MODEL_CONFIG_PATH] [-t TRAIN_CONFIG_PATH] [--vocoder_checkpoint VOCODER_CHECKPOINT_PATH] [--vocoder_config VOCODER_CONFIG_PATH]

TensorBoard

Use

tensorboard [--logdir LOG_PATH]

• Tensorboard for pretrain model

• Tensorboard for finetune with only 5 sentences

References

• AdaSpeech: Adaptive text to speech for custom voice.
• ming024's implementation
• rishikksh20's AdaSpeech implementation

Citation

@misc{chen2021adaspeech,
      title={AdaSpeech: Adaptive Text to Speech for Custom Voice}, 
      author={Mingjian Chen and Xu Tan and Bohan Li and Yanqing Liu and Tao Qin and Sheng Zhao and Tie-Yan Liu},
      year={2021},
      eprint={2103.00993},
      archivePrefix={arXiv},
      primaryClass={eess.AS}
}