Latent source-specific generative factor learning for monaural speech separation using weighted-factor autoencoder

• Published in: Frontiers of information technology & electronic engineering Vol. 21; no. 11; pp. 1639 - 1650
• Main Authors: Chen, Jing-jing, Mao, Qi-rong, Qin, You-cai, Qian, Shuang-qing, Zheng, Zhi-shen
• Format: Journal Article
• Language: English
• Published: Hangzhou Zhejiang University Press 01.11.2020; Springer Nature B.V; School of Computer Science and Communication Engineering, Jiangsu University, Zhenjiang 212013, China%School of Computer Science and Communication Engineering, Jiangsu University, Zhenjiang 212013, China; Jiangsu Key Laboratory of Security Technology for Industrial Cyberspace, Zhenjiang 212013, China
• Subjects: Communications Engineering; Computer Hardware; Computer Science; Computer Systems Organization and Communication Networks; Deep learning; Dictionaries; Electrical Engineering; Electronics and Microelectronics; Instrumentation; Microphones; Networks; Neural networks; Regularization; Separation; Speech; Deep learning; Speech separation; TN912.3; Generative factors; Autoencoder
• ISSN: 2095-9184; 2095-9230
• DOI: 10.1631/FITEE.2000019

Much recent progress in monaural speech separation (MSS) has been achieved through a series of deep learning architectures based on autoencoders, which use an encoder to condense the input signal into compressed features and then feed these features into a decoder to construct a specific audio source of interest. However, these approaches can neither learn generative factors of the original input for MSS nor construct each audio source in mixed speech. In this study, we propose a novel weighted-factor autoencoder (WFAE) model for MSS, which introduces a regularization loss in the objective function to isolate one source without containing other sources. By incorporating a latent attention mechanism and a supervised source constructor in the separation layer, WFAE can learn source-specific generative factors and a set of discriminative features for each source, leading to MSS performance improvement. Experiments on benchmark datasets show that our approach outperforms the existing methods. In terms of three important metrics, WFAE has great success on a relatively challenging MSS case, i.e., speaker-independent MSS.