Abstract: Monaural Speech Segregation Based on Pitch Track Correction Using Particle Filtering

We propose a novel method to improve the performance of monaural speech segregation. Our method is based on pitch track correction using particle filtering. The conventional approach for pitch track correction simply uses the peaks of the autocorrelation functions, depending only on the longest reliable pitch streak. To enhance the conventional approach, we consider all reliable pitch streaks rather than only the longest one and correct unreliable pitch using a particle filter. To apply the particle filtering technique for pitch track correction, an importance weight computation method using foreground stream is proposed. To verify the efficiency of the proposed method, a number of speech segregation experiments for mixtures of speech and various competing sound sources in various mixing signal-to-noise ratios (SNRs) were carried out. With respect to several performance measures including SNR, energy loss ratio, and noise residue ratio of the segregated speech, the proposed method exhibits better performance than the conventional approach.