Blind Assessment of Microphone Utility in a Wireless Acoustic Sensor Network

In a wireless acoustic sensor network where microphones are distributedly scattered, the microphone subset selection is usually based on the quality of the received signals on the microphones. However, it is difficult to evaluate it without knowing about the nodes and the sound sources position, the source signal and the noise signals. In this paper, a novel algorithm is proposed depending only on the received data for assessing microphone utility blindly. The algorithm exploits the relationship between the SNR and the signals higher-order statistical information (kurtosis) in frequency domain and the microphone utility is a weighted sum of the kurtosis of each frequency bin. The experimental results show that the proposed algorithm can effectively evaluate the quality of the received signals and get the comparable performance with the true SNR.