TraceableSpeech

PyTorch Implementation of TraceableSpeech: Towards Proactively Traceable Text-to-Speech with Watermarking

Now we open the part of speech watermarking.

This is the watermark training pipeline.

Quick Started

Dependencies

pip install -r requirement.txt

Default Preparation

We are using the LibriTTS dataset.

Modify the parameter --input_training_file --input_validation_file --checkpoint_path in train.py

Modify the parameter --input_wavs_dir --output_dir --checkpoint_file in inference.py

Modify the config.json

Watermark config

The watermark configuration is in the watermark.py file, defaulting to 4-digit base-16.

Train

python train.py

Test

python inference.py

Acknowledgements

This implementation uses parts of the code from the following Github repos: AcademiCodec

Citations

If you find this code useful in your research, please consider citing:

@misc{zhou2024traceablespeechproactivelytraceabletexttospeech,
      title={TraceableSpeech: Towards Proactively Traceable Text-to-Speech with Watermarking}, 
      author={Junzuo Zhou and Jiangyan Yi and Tao Wang and Jianhua Tao and Ye Bai and Chu Yuan Zhang and Yong Ren and Zhengqi Wen},
      year={2024},
      eprint={2406.04840},
      archivePrefix={arXiv},
      primaryClass={cs.SD},
      url={https://arxiv.org/abs/2406.04840}, 
}