Simplified pitch detection algorithm of mixed speech signals

In the speech processing system, detection of a single speech from composite signals is required. To obtain the single speech, we propose an algorithm which can detect multi-pitch in the frame processing. The algorithm does not require the condition that the overlapped peaks should be analyzed into each speaker's peaks. The proposed algorithm produces pitch candidates for each speech such that the pitches are not correlated and are accurate below the FFT frequency unit. We show that the proposed algorithm works properly in the case of two speaker speech signals.