Universal Speech Enhancement Demo

A single, universal speech enhancement (SE) model effectively addresses diverse distortions—including additive noise, reverberation, clipping, and bandwidth limitations—across a full range of supported sampling frequencies (8 kHz to 48 kHz). All demo samples below are produced by this single model.

1. Additive Noise

Noisy Sample (fileid_13)

Enhanced Sample (fileid_13)

Noisy Sample (fileid_14)

Enhanced Sample (fileid_14)

Noisy Sample (fileid_17)

Enhanced Sample (fileid_17)

Noisy Sample (fileid_549)

Enhanced Sample (fileid_549)

Noisy Sample (fileid_690)

Enhanced Sample (fileid_690)

Noisy Sample (fileid_834)

Enhanced Sample (fileid_834)

Noisy Sample (fileid_975)

Enhanced Sample (fileid_975)

2. Clipping

Noisy Sample (fileid_3)

Enhanced Sample (fileid_3)

Noisy Sample (fileid_20)

Enhanced Sample (fileid_20)

Noisy Sample (fileid_28)

Enhanced Sample (fileid_28)

Noisy Sample (fileid_45)

Enhanced Sample (fileid_45)

Noisy Sample (fileid_735)

Enhanced Sample (fileid_735)

Noisy Sample (fileid_986)

Enhanced Sample (fileid_986)

3. Reverberation

Noisy Sample (fileid_22)

Enhanced Sample (fileid_22)

Noisy Sample (fileid_843)

Enhanced Sample (fileid_843)

Noisy Sample (fileid_995)

Enhanced Sample (fileid_995)

Noisy Sample (fileid_999)

Enhanced Sample (fileid_999)

4. Bandwidth Limitation (working on it to further improve the performance)

Noisy Sample (fileid_21)

Enhanced Sample (fileid_21)

Noisy Sample (fileid_57)

Enhanced Sample (fileid_57)

Noisy Sample (fileid_152)

Enhanced Sample (fileid_152)

Citation:

If you find the results on this demo useful in your research, please reference our foundational paper:

An Investigation of Incorporating Mamba for Speech Enhancement

@article{chao2024investigation,
    title = {An Investigation of Incorporating Mamba for Speech Enhancement},
    author = {Chao, Rong and Cheng, Wen-Huang and La Quatra, Moreno and Siniscalchi, Sabato Marco},
    journal = {arXiv preprint arXiv:2405.06573},
    year = {2024}
}