Voice activity detector based on pitch statistics for speaker recognition

A modification of the voice activity detection (VAD) based on pitch statistics is proposed for improving speaker recognition performance. Comparison of two approaches for the VAD choice for phonogram preprocessing is presented. The first approach is a modification of pitch statistics VAD, the second one represents the simple energy-based scheme. A quantitative measurement of verification system quality is also presented. The experimental results show that the proposed VAD essentially improves the performance of the speaker recognition systems.