Derivative of instantaneous frequency for voice activity detection using phase-based approach

In this paper, we consider the use of the phase spectrum in speech signal analysis. In particular, a phase-based voice activity detection (VAD) by using the derivative of instantaneous frequency is proposed. Preliminary experiments reveal that the distribution of this feature can indicate the presence or absence of speech. The performance of the proposed method is evaluated in comparison with the conventional amplitude-based method. In addition, we consider a combination of the amplitude-based and phase-based methods in a simple manner to demonstrate the complementarity of both spectra. The experimental results confirm that the phase information can be used to detect voice activity with at least 62% accuracy. The proposed method shows better performance compared to the conventional amplitude-based method in the case when a speech signal was corrupted by white noise at low signal-to-noise ratio (SNR). A combination of two methods achieves even higher performance than each of them separately, in limited conditions.