StyleSinger: Style Transfer for Out-of-Domain Singing Voice Synthesis

Yu Zhang, Rongjie Huang, Ruiqi Li, JinZheng He, Yan Xia, Feiyang Chen, Xinyu Duan, Baoxing Huai, Zhou Zhao | Zhejiang University, Huawei Cloud

PyTorch Implementation of StyleSinger (AAAI 2024): Style Transfer for Out-of-Domain Singing Voice Synthesis.

We provide our implementation and pre-trained models in this repository.

Visit our demo page for audio samples.

News

• 2024.09: We released the full dataset of GTSinger!
• 2024.05: We released the code and checkpoints of StyleSinger!
• 2023.12: StyleSinger is accepted by AAAI 2024!

Key Features

• We present StyleSinger, the first singing voice synthesis model for zero-shot style transfer of out-of-domain reference samples. StyleSinger excels in generating exceptional singing voices with unseen styles derived from reference singing voice samples.
• We propose the Residual Style Adaptor (RSA), which uses a residual quantization model to meticulously capture diverse style characteristics in reference samples.
• We introduce the Uncertainty Modeling Layer Normalization (UMLN) to perturb the style information in the content representation during the training phase, and thus enhance the model generalization of StyleSinger.
• Extensive experiments in zero-shot style transfer show that StyleSinger exhibits superior audio quality and similarity compared with baseline models.

Quick Start

We provide an example of how you can generate high-fidelity samples using StyleSinger.

To try on your own dataset or GTSinger, simply clone this repo on your local machine provided with NVIDIA GPU + CUDA cuDNN and follow the below instructions.

Pre-trained Models

You can use all pre-trained models we provide on HuggingFace or Google Drive. Notably, this StyleSinger checkpoint only supports Chinese! You should train your own model based on GTSinger for multilingual style transfer! Details of each folder are as follows:

Model	Description
StyleSinger	Acousitic model (config)
HIFI-GAN	Neural Vocoder
Encoder	Emotion Encoder

Dependencies

A suitable conda environment named stylesinger can be created and activated with:

conda create -n stylesinger python=3.8
conda install --yes --file requirements.txt
conda activate stylesinger

Multi-GPU

By default, this implementation uses as many GPUs in parallel as returned by torch.cuda.device_count(). You can specify which GPUs to use by setting the CUDA_DEVICES_AVAILABLE environment variable before running the training module.

Inference for Chinese singing voices

Here we provide a speech synthesis pipeline using StyleSinger.

1. Prepare StyleSinger (acoustic model): Download and put checkpoint at checkpoints/StyleSinger.
2. Prepare HIFI-GAN (neural vocoder): Download and put checkpoint at checkpoints/hifigan.
3. Prepare Emotion Encoder: Download and put checkpoint at checkpoints/global.pt.
4. Prepare reference information: Provide a reference_audio (48k) and input target ph, target note for each ph, target note_dur for each ph, target note_type for each ph (rest: 1, lyric: 2, slur: 3), and reference audio path. Input these information in Inference/StyleSinger.py. Notably, if you want to use Chinese data in GTSinger to infer this Chinese checkpoint, refer to phone_set, you have to delete _zh in each ph of GTSinger, and change <AP> to breathe, <SP> to _NONE!
5. Infer for style transfer:

rm -rf ./checkpoints/checkpoints
CUDA_VISIBLE_DEVICES=$GPU python inference/StyleSinger.py --config egs/stylesinger.yaml  --exp_name checkpoints/StyleSinger

Generated wav files are saved in infer_out by default.

Train your own model based on GTSinger

Data Preparation

1. Prepare your own singing dataset or download GTSinger.
2. Put metadata.json (including ph, word, item_name, ph_durs, wav_fn, singer, ep_pitches, ep_notedurs, ep_types for each singing voice) and phone_set.json (all phonemes of your dictionary) in data/processed/style (Note: we provide metadata.json and phone_set.json in GTSinger, but you need to change the wav_fn of each wav in metadata.json to your own absolute path).
3. Set processed_data_dir (data/processed/style), binary_data_dir, valid_prefixes (list of parts of item names, like ["Chinese#ZH-Alto-1#Mixed_Voice_and_Falsetto#一次就好"]), test_prefixes in the config.
4. Download the global emotion encoder to emotion_encoder_path (training on Chinese only) or train your own global emotion encoder referring to Emotion Encoder based on emotion annotations in GTSinger.
5. Preprocess Dataset:

export PYTHONPATH=.
CUDA_VISIBLE_DEVICES=$GPU python data_gen/tts/bin/binarize.py --config egs/stylesinger.yaml

Training StyleSinger

CUDA_VISIBLE_DEVICES=$GPU python tasks/run.py --config egs/stylesinger.yaml  --exp_name StyleSinger --reset

Inference using StyleSinger

CUDA_VISIBLE_DEVICES=$GPU python tasks/run.py --config egs/stylesinger.yaml  --exp_name StyleSinger --infer

Acknowledgements

This implementation uses parts of the code from the following Github repos: GenerSpeech, NATSpeech, ProDiff, DiffSinger as described in our code.

Citations

If you find this code useful in your research, please cite our work:

@inproceedings{zhang2024stylesinger,
  title={StyleSinger: Style Transfer for Out-of-Domain Singing Voice Synthesis},
  author={Zhang, Yu and Huang, Rongjie and Li, Ruiqi and He, JinZheng and Xia, Yan and Chen, Feiyang and Duan, Xinyu and Huai, Baoxing and Zhao, Zhou},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={38},
  number={17},
  pages={19597--19605},
  year={2024}
}

Disclaimer

Any organization or individual is prohibited from using any technology mentioned in this paper to generate someone's singing without his/her consent, including but not limited to government leaders, political figures, and celebrities. If you do not comply with this item, you could be in violation of copyright laws.