Pitch Determination of Noisy Speech Using Cumulant Based Modified Weighted Function

In this paper, we propose a pitch extraction method effective in highly noisy environments. The proposed method utilizes the cumulant based modified autocorrelation function (CMACF) and cumulant based circular average magnitude difference function (CCAMDF) in weighted autocorrelation function (WAF), both of which present the periodicity at the pitch period of voiced speech. The CMACF exhibits a peak while the CCAMDF produces a notch. The modified weighted function is developed by the multiplication of the amplitude spectrum of CMACF and the inverse of the characteristics of CCAMDF, to enhance the accuracy of pitch extraction. Experimental results show that the proposed method provides a superior performance than the conventional methods under lower levels of signal-to-noise ratio in white noise.