UMP Test Based Voice Activity Detection Using High Frequency Resolution

This paper presents a new high frequency resolution framework to improve uniformly most powerful (UMP) test based voice activity detection (VAD). The high frequency resolution is employed to make use of the contextual information of neighbor frames. The VAD decision is made by comparing the geometric average of SNR measures to the adaptive threshold which is decided with noise statistics by UMP test. Cepstrum thresholding method is also used for noise spectrum estimation instead of Welch method. Experimental results show that the proposed VAD algorithm based on high frequency resolution outperforms the normal scheme at all SNR scenarios.