A residual-cepstrum method of pitch estimation from noisy speech
Celia Shahnaz, Wei‐Ping Zhu, M. Omair Ahmad · Canadian acoustics · 2009
Residual and cepstral representations of speech were utilized to estimate pitch in a noisy environment. It was found that the major excitation of the vocal tract within a pitch period occurred at the instant of glottal closure (GC). It was possible to determine the pitch period by careful analysis of the speech signal with the help of GC instants. A discrete Fourier transform (DFT) based power cepstrum (DFTPC) was proposed to overcome the adverse effect of noise on the Hilbert envelope (HE) and residual signal (RS). The DFTPC of the HE exhibited a more prominent pitch-peak in a heavily degraded condition in comparison to that demonstrated by the conventional cepstrum of the noisy speech. Simulation results indicated that the global maximization of the DFTPC yielded an accurate pitch estimate as compared to the latest models for a wide range of speakers in noisy environments.