Intelligibility Enhancement Through Time-Frequency Masking with Distributed Microphone System

The proposed work focuses on continued development of masking algorithms to extract weak voices in cocktail party environments using distributed microphones. This builds on previous work where a toolbox was developed and sent to sponsor for detecting and locating sources, beamforming, and masking for intelligibility enhancement with distribute microphones. The next year’s funding will primarily support a graduate student to improve masking algorithm performance and develop more efficient implementations, which may include using with wavelet packets in the time domain. Experiments based on real recordings and simulations will be performed for testing performance. Best performing algorithms will be packaged in graphical user interfaces and toolbox function for use by the sponsor.