Gabor Filters Based Segmentation For Monaural Sound Segregation

Monaural sound segragation has been a very challenging problem where blind source separation techniques could not directly be applied. This study proposes a new time-frequency spectrogram segmentation technique for segregation of human speech from monaural sound mixtures. Gabor filters are used to spectrogram image to enhance segment contours, then segmentation process is applied. The system is tested on speech signals from TIMIT database, mixed with several noise sources with SNR ration up to 0 dB, and encouraging results are obtained.