A new perceptual convolutive blind source separation algorithm for speech separation

• Published in: 2004 7th International Conference on Signal Processing : proceedings : ICSP'04 : Aug. 31-Sept 4, 2004, Beijing, China Vol. 1; pp. 323 - 326 vol.1
• Main Authors: Qiongfeng Pan, Aboulnasr, T.
• Format: Conference Proceeding
• Language: English
• Published: Beijing IEEE 2004; Piscataway NJ Pub. House of Electronics Industry
• Subjects: Applied sciences; Auditory system; Blind source separation; Detection, estimation, filtering, equalization, prediction; Ear; Exact sciences and technology; Frequency; Humans; Information, signal and communications theory; Least squares approximation; Miscellaneous; Psychoacoustic models; Signal and communications theory; Signal processing; Signal, noise; Source separation; Speech; Telecommunications and information theory; Performance evaluation; Simulation; Signal processing; Masking; Blind source separation; Algorithm; Least mean squares methods
• ISBN: 9780780384064; 0780384067
• DOI: 10.1109/ICOSP.2004.1452647

In this paper, we address the problem of blind speech signal separation. A new perceptual convolutive blind source separation algorithm is proposed based on filtered-E LMS algorithm and the masking properties of human auditory system. This algorithm emphasizes frequencies to which the human ear is sensitive and de-emphasizes frequencies that are inaudible to human ear by incorporating the properties of human auditory system making it suitable for speech separation. Simulation results on blind speech signal separation show that the proposed algorithm can improve the separated speech quality and has better convergence performance.