Automatic adaptive speech separation using beamformer-output-ratio for voice activity classification
Date
2015
Authors
Tran, T.N.
Cowley, W.
Pollok, A.
Editors
Advisors
Journal Title
Journal ISSN
Volume Title
Type:
Journal article
Citation
Signal Processing, 2015; 113:259-272
Statement of Responsibility
Conference Name
Abstract
This paper focuses on the practical challenge of adaptation control for speech separation systems. Adaptive beam forming methods, such as minimum variance distortion less response (MDVR), can effectively extract the desired speech signal from interference and noise. However, to avoid the signal cancellation problem, the beamformer adaptation is halted when the desired speaker is active. An automated scheme for this adaptation requires classifying speakers' voice activity status, which remains a challenge for multispeaker environments. In this paper, we propose a novel approach to identify voice activities for two speakers based on a new metric, called the beam former-output-ratio (BOR). Statistical properties of the BOR are studied and used to develop a hypothesis based method for voice activity classification. The method is further refined using an algorithm detecting incorrect beamformer adaptation by analysing changes in the output power of a blind adapting MVDR beamformer. Based on the new methods, we constructan automatic adaptive beamforming system to simultaneously separate speech for two speakers. The speech separation module of the system uses MVDR beamformers whose adaptation is guided by the voice activity classification. Our methods can lead to, in some cases, 20% reduction in voice activity classification error, and 8 dB improvement on theoutput SINR. The results are verified on both synthesised signals and realistic recordings.
School/Discipline
Dissertation Note
Provenance
Description
Access Status
Rights
Copyright 2015 Elsevier