Separation of voice and musical sound in wavelet domain using harmonically nested microphone array

摘要

We propose a new method for enhancement of speech signal interfered with musical sounds, using harmonically nested microphone array and wavelet packet filtering. The system is constructed from two parts called "array signal processing" and "separation & removal of musical signal" respectively. In the "array signal processing" two different processes, "delay-and-sum beamforming" and "blocking process” that both separate signals spatially, are performed for the received signals in each microphone. In the "separation & removal of musical signal", subbands which contain musical sounds mainly are removed in the wavelet domain on the basis of correlation coefficients between the two outputs of array processing part. Furthermore, improving the spatial resolution of the microphone array, we adopt the harmonic nesting as one of more suitable sensor arrangement for the broadband signals.