Results
We simulated a realistic noisy environment by mixing speech samples with restaurant background noise. The system successfully reduced the noise and improved overall speech quality. The system was most effective in highly noisy environments with improvements of up to 15 dB in SNR. In addition, the real-time processing delay was less than a 35 ms, which is within television standards for lip sync error [5].
Future Work
There are several possible extensions and improvements to our speech enhancement system. For example, other beamforming techniques, such as the super-directive beamformer, could reduce the physical size of the system and improve directivity [2]. In addition, using stochastic noise cancellation algorithms might make the system more robust and improve performance in quiet environments [6].