Determination of pitch of noisy speech using dominant harmonic frequency

This paper presents a noise robust method for determination of pitch using least-squares estimate of the dominant harmonic frequency (DHF) from observations of speech, represented by a harmonic sine-wave model, heavily corrupted by white noise. The DHF is effectively estimated using the proposed sinusoidal autocorrelation model (SAM) of the noise-free speech signal. The method is based on matching the period of an impulse train, varied as the integer multiple of the estimated DHF, with the enhanced autocorrelation function of the noisy speech emphasized by the average magnitude difference function (AMDF). The simulation results show that the proposed method can detect pitch with higher accuracy even at a signal to noise ratio (SNR) as low as -5 dB.