듀얼 마이크를 이용한 교차 파워 스펙트럼 위상 기반의 음성 구간 검출 알고리즘

Abstract

To detect voice activity in adverse noisy environment, multichannel signal processing can be considered to exploit spatial selectivity. In this paper, a dual-microphone voice activity detection method is proposed to cope with the presence of strong coherent interference. Under the assumption that the direction of the target speech signal is known and the time delay between the two microphones is compensated, it is shown that the maximum of windowed averaging of real coherence function provides a reliable feature for the voice activity detection even in low signal-to-interference ratio case. The coherence function is based on the cross power spectrum phase between the two microphones. In the experiments, isolated digit utterances are convolved with impulse responses to simulate dual-microphone signals. The experimental results show that the proposed method provides more reliable feature for the voice activity detection in the presence of strong coherent interference.