New speech enhancement techniques using the pitch mode modulation model

Speech enhancement is often required prior to speech recognition and speaker identification (SI) systems. The new speech enhancement techniques described herein are based on the Pitch Mode Modulation Model (PMMM). The PMMM exploits the time-synchrony present in speech to time-warp the speech waveform such that the quasi-periodic signal becomes periodic. After this nonlinear preprocessing step, the speech signal is enhanced by comb filtering and rewarping. The enhancement technique is used with a standard cepstrum-based SI system and is shown to significantly reduce the sensitivity of the system to mismatches between training and testing data. Other applications of the enhancement model are also discussed