A two-sensor voice activity detection and speech enhancement based on coherence with additional enhancement of low frequencies using pitch information
Alexandre Guérin · European Signal Processing Conference · 2000
This report proposesa2 microphone VoiceActivityDetector(VAD)and a Speech Enhancer (ENH) adapted to car conditions. The two modules are derived from the well-known Magnitude Square Coherence (MSC) which expresses a normalized cross-correlation for each frequency band of the received signals by the two sensors. A global VAD is directly obtained from the MSC by adaptive threshold which ensures a quasi-constant behaviour in different environmental conditions and different relative microphones positions. The ENH filter is applied to one of the two microphones and is divided in two parts : 1. from a Modified Coherence Function including Power Cross-Spectral Subtraction of background noise estimation is based on the Wiener Filter that is used to enhance speech; the estimation of Power Spectral Densities (PSD) are optimized to prevent the emergence of musical noise as well as reverberant effect. 2. a second module extracts the pitch value of voiced sections of speech to enhance low frequency bands of main signal that are partially or even totally removed by Wiener Filtering.