4 This paper describes an approach used to separate overlapping speech signals. This approach is comprised of overlapping speech analysis stage and waveform synthesis stage. In the speech analysis stage, we use two ways to extract pitch flow, one way is based on SHS algorithm, SHS on single speech algorithm is modified, pitch梩racking and pitch梕xtraction process are incorporated into a closedæ¢oop system. The other way is based on wavelet transform, we use wavelet transform to fulfil the same task but in a different way. Both the ways based on the concept that both glottal excitation and vocal tract parameters are time varying in amplitude and frequency. To track this tendency we propose the concept of pitchæ¢ikelihood梖unction. In synthesis stage, using sinusoidal model, a set of amplitudes, frequencies and phases of harmonics based on pitch flow getting from analysis stage act as the parameters of sinusoidal model, and generate the synthetic speech waveform. In order to make the frame edge maximally smooth, we propose the matching concept. These matches include frequencies matching, amplitudes interpolating, phases interpolating. The results obtained indicate that the approach described here is efficient. ZhuMe iHong Directed by ZhaoHeMing...
|