Multi-Pitch Detection for Co-Channel Speech Utilizing Frequency Channel Piecewise Integration and Morphological Feedback Verification Tracking

We propose an effective algorithm to detect the multi-pitch of co-channel speech with two speakers. A discriminative scheme is exploited to gather harmonic information according to the piecewise continuity of autocorrelation peaks across different frequency channels within a time frame. A group of adaptive morphological filters and feedback verification are employed to coordinate and verify the periodicity information across time frames. Quantitative evaluation shows that the proposed system gains excellent results in veracity and integrality comparing with existing good algorithms of multi-pitch detection for co-channel speech. In addition, the proposed algorithm performs efficiently in timeconsuming as well as memory-consuming.