Noise estimation using negentropy based voice-activity detector

This paper presents a noise robust voice activity detector (VAD) based on the chaos measure of the quasistationary segment of speech in the frequency domain. The basic idea behind the proposed method is that the addition of noise in the clean speech produces less disorganization in the speech part than the silent/paused part as a result of which spectral contents of the speech part become less chaotic than the silent part in a noisy speech signal. The negentropy can be used as the measure for such a distinction in organization and accordingly noisy speech frames and the noise-only frames are demarcated. The algorithm is almost independent of the SNR level of the noisy speech signal.