Study On Distributed Speech Enhancement In Wireless Acoustic Sensor Networks

Driven by a wide of speech processing applications, speech enhancement technology, which can extract the desired speech signal from the noisy speech signal, has always been a research focus. In recent years, with the rise of sensor networks, wireless acoustic sensor networks (WASNs) are rapidly developed and applied to many research areas, such as military and environmental monitoring. A WASNs can be used to perceive and monitor the external environment, calculate and process data, and the nodes in the WASNs can communicate with other nodes to achieve information sharing and collaboration between nodes. Thus, how to use the advantages of WASNs for distributed speech enhancement is also a novel research hotspot.In this thesis speech enhancement methods using the traditional microphone array and the features and functions of WASNs are conducted in-depth research and discussion. And, based on the above research findings, two distributed speech enhancement algorithm are proposed. The main works of this thesis are as follows:(1) distributed speech enhancement algorithm based on distributed consistency and minimum variance distortionless response (MVDR). The centralized MVDR algorithm is decomposed into each node to achieve a MVDR data preprocessing using the recording sound source data. Then, based on the average Metropolis weights, a linear distributed averaging consensus iteration algorithm is conducted to achieve a consistent speech enhancement for each node. Experimental results show that the proposed method can effectively suppress interference caused by the incoherent noise, and each node can obtain an enhanced speech approaching to the SNR of the sound source signal.(2) distributed least mean-square (LMS) for consensus-based speech enhancement algorithm. Based on the minimum mean square error criterion and an neighboring nodes communication method, the algorithm uses a consistent LMS iteration process to cope with the received noisy speech data to achieve a distributed speech enhancement. Experimental results show that the method can effectively suppress the interference caused by coherent noise, and the enhanced signal of each node can approach to the SNR of the sound source.