Speech enhancement in wireless acoustic sensor networks (SEWASN)

Themes: Audio and acoustic signal processing

To improve the quality and intelligibility of speech processing applications in noisy environments, it is customary to equip them with a single- or multi-microphone noise reduction algorithm. As multi-microphone noise reduction algorithms can take advantage of spatial diversity, they usually lead to better speech quality and intelligibility than their single-microphone counterparts. It is in particular the number of microphones and their placement that determine the potential performance of a multi-microphone noise reduction algorithm. However, as most mobile speech processing devices have relatively small dimensions, the number of microphones as well as their placement is rather restricted.

Using so-called wireless acoustic sensor networks (WASNs), it is possible to use a much larger number of microphones that are distributed in the environment and where their placement is not restricted by the device itself. This allows a further increase in noise reduction performance. However, the conventional multi-microphone noise reduction algorithms are characterized by having one processor where all data is processed centrally. In this project, the goal is to develop distributed speech enhancement algorithms for wireless acoustic sensor networks.