A Generalized Subspace Approach for Multichannel Speech Enhancement Using Machine Learning-Based Speech Presence Probability Estimation

Ke Y, Hu Y, Li J, Zheng C, Li X. A Generalized Subspace Approach for Multichannel Speech Enhancement Using Machine Learning-Based Speech Presence Probability Estimation. In: AES Convention 146. Audio Engineering Society; 2019. Paper 10192. Available from: https://aes.org/publications/elibrary-page/?id=20325

@inproceedings{Ke2019_20325,
  author = {Ke, Yuxuan and Hu, Yi and Li, Jian and Zheng, Chengshi and Li, Xiaodong},
  title = {{A Generalized Subspace Approach for Multichannel Speech Enhancement Using Machine Learning-Based Speech Presence Probability Estimation}},
  booktitle = {AES Convention 146},
  note = {Paper 10192},
  year = {2019},
  month = mar,
  publisher = {Audio Engineering Society},
  url = {https://aes.org/publications/elibrary-page/?id=20325}
}

TY  - CPAPER
TI  - A Generalized Subspace Approach for Multichannel Speech Enhancement Using Machine Learning-Based Speech Presence Probability Estimation
AU  - Ke, Yuxuan
AU  - Hu, Yi
AU  - Li, Jian
AU  - Zheng, Chengshi
AU  - Li, Xiaodong
T2  - AES Convention 146
M1  - Paper 10192
PY  - 2019
DA  - 2019/03/06
UR  - https://aes.org/publications/elibrary-page/?id=20325
PB  - Audio Engineering Society
LA  - en
AB  - A generalized subspace-based multichannel speech enhancement in frequency domain is proposed by estimating multichannel speech presence probability using machine learning methods. An efficient and low-latency neural networks (NN) model is introduced to discriminatively learn a gain mask for separating the speech and the noise components in noisy scenarios. Besides, a generalized subspace-based approach in frequency domain is proposed, where the speech power spectral density (PSD) matrix and the noise PSD matrix are estimated by short-term and long-term averaging periods, respectively. Experimental results show that the proposed method outperforms the existing NN-based beamforming methods in terms of the perceptual evaluation of speech quality score and the segmental signal-to-noise ratio improvement.
ER  -