Phase-aware Speech Enhancement with
Deep Complex Unet

Evaluation Result

Quantitative evaluation results of other algorithms and our method.

CSIG CBAK COVL PESQ SSNR
Wiener(Scalart et al., 1996) 3.23 2.68 2.67 2.22 5.07
SEGAN (Pascual et al., 2017) 3.48 2.94 2.80 2.16 7.73
Wavenet (Rethage et al. 2017) 3.62 3.23 2.98 None None
MMSE-GAN(Soni et al.) 3.80 3.12 3.14 2.53 None
Deep Feature Loss (Germain et al., 2018) 3.86 3.33 3.22 None None
DCUnet-20 4.24 4.00 3.69 3.13 15.95
Large-DCUnet-20 4.34 4.10 3.81 3.24 16.85

Audio Samples


SNR Sample Name Mixture Clean DeepComplexUNet
2.5dB 232_052
232_170
257_054
257_070
257_143
SNR Sample Name Mixture Clean DeepComplexUNet
7.5dB 232_013
232_095
232_103
232_266
257_154
SNR Sample Name Mixture Clean DeepComplexUNet
12.5dB 232_306
232_367
257_289
257_329
257_432
SNR Sample Name Mixture Clean DeepComplexUNet
17.5dB 257_140
257_252
257_308
257_388
257_408

The audio samples of other comparable algorithms are available in the links below.
We encourage listeners to use headphone for more precise comparison with other algorithms.

SEGAN     Wavenet Denoising     Deep Feature Loss

References


[1] Chiheb Trabelsi, Olexa Bilaniuk, Ying Zhang, Dmitriy Serdyuk, Sandeep Subramanian, Joao Felipe Santos, Soroush Mehri, Negar Rostamzadeh, Yoshua Bengio, and Christopher J Pal. Deep complex networks. arXiv preprint arXiv:1705.09792, 2017.

[2] Pascal Scalart et al. Speech enhancement based on a priori signal to noise estimation. In Acoustics, Speech, and Signal Processing, 1996. ICASSP-96. Conference Proceedings., 1996 IEEE International Conference on, volume 2, pp. 629–632. IEEE, 1996.

[3] Santiago Pascual, Antonio Bonafonte, and Joan Serra. Segan: Speech enhancement generative adversarial network. arXiv preprint arXiv:1703.09452, 2017.

[4] Dario Rethage, Jordi Pons, and Xavier Serra. A wavenet for speech denoising. arXiv preprint arXiv:1706.07162, 2017.

[5] Meet H Soni, Neil Shah, and Hemant A Patil. Time-frequency masking-based speech enhancement using generative adversarial network. In IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2018.

[6] Francois G Germain, Qifeng Chen, and Vladlen Koltun. Speech denoising with deep feature losses. arXiv preprint arXiv:1806.10522, 2018.