Interspeech'2005 - Eurospeech

Lisbon, Portugal
September 4-8, 2005

A Study of Weighted CSP Analysis with Average Speech Spectrum for Noise Robust Talker Localization

Yuki Denda, Takanobu Nishiura, Yoichi Yamashita

Ritsumeikan University, Japan

This paper describes a new method of noise robust talker localization for the front-end processing of microphone array steering. Conventional talker localization methods cannot localize a target talker accurately in higher noisy environments. To deal with this problem, in this paper, we propose an weighted CSP (Cross-power Spectrum Phase) analysis with an average speech spectrum. The proposed method consists of two processes. At first, CSP coefficients are weighted by analysis weight coefficients based on average speech spectrum, which is trained with speech database, in advance. Next, the interference noises are reduced on spatial domain by CSP coefficients subtraction. As a result of evaluation experiments in a real room, we confirmed that the proposed method could provide better talker localization performance than the conventional methods.

Full Paper

Bibliographic reference.  Denda, Yuki / Nishiura, Takanobu / Yamashita, Yoichi (2005): "A study of weighted CSP analysis with average speech spectrum for noise robust talker localization", In INTERSPEECH-2005, 2321-2324.