[2024-05-09] Update SPMamba WHAM! Result: SI-SNRi=17.4 dB, SDRi=17.6 dB

[2024-04-23] Update SPMamba MACs: 238.21 G/s using code

[2024-04-18] Update SPMamba WSJ0-2Mix Result: SI-SNRi=22.5 dB, SDRi=22.7 dB

SPMamba revolutionizes the field of speech separation tasks by leveraging the power of Mamba in conjunction with the robust TF-GridNet infrastructure. By replacing the conventional bidirectional LSTM with a more efficient and effective bidirectional Mamba model, SPMamba sets a new standard for accuracy and performance in speech separation.

Built upon the open-source ESPnet framework, SPMamba offers a seamless experience for users looking to train their models with cutting-edge technology. Whether you're a researcher, developer, or enthusiast in the field of speech processing, SPMamba provides the tools and flexibility needed to achieve unparalleled results.

This repository is implemented using the ESPnet framework, a comprehensive platform for speech processing. SPMamba enhances ESPnet by integrating Mamba, a state-of-the-art state-space model, into the TF-GridNet architecture. This combination allows for significant improvements in speech separation tasks.

clone the repository

git clone https://github.com/JusperLee/SPMamba.git && cd SPMamba
conda env create -f look2hear.yml
conda activate look2hear

To train the SPMamba model, run the following command:

python audio_train.py --conf_dir=configs/spmamba.yml

Here, you can include a brief overview of the performance metrics or results that SPMamba achieves using own private datasets.

SPMamba is licensed under the Apache License 2.0. For more details, see the LICENSE file in the repository.

SPMamba is developed by the Look2Hear team at Tsinghua University. We would like to thank the ESPnet team for their contributions to the open-source community and for providing a solid foundation for our work.

If you use SPMamba in your research or project, please cite the following paper:

@article{li2024spmamba,
  title={SPMamba: State-space model is all you need in speech separation},
  author={Li, Kai and Chen Guo},
  journal={arXiv preprint arXiv:2404.02063},
  year={2024}
}

For any questions or feedback regarding SPMamba, feel free to reach out to us via email: [email protected]