Formant frequency estimation of high-pitched vowels using weighted linear prediction

Abstract

All-pole modeling is a widely used formant estimation method, but its performance is known to deteriorate for high-pitched voices. In order to address this problem, several all-pole modeling methods robust to fundamental frequency have been proposed. This study compares five such previously known methods and introduces a technique, Weighted Linear Prediction with Attenuated Main Excitation (WLP-AME). WLP-AME utilizes temporally weighted linear prediction (LP) in which the square of the prediction error is multiplied by a given parametric weighting function. The weighting downgrades the contribution of the main excitation of the vocal tract in optimizing the filter coefficients. Consequently, the resulting all-pole model is affected more by the characteristics of the vocal tract leading to less biased formant estimates. By using synthetic vowels created with a physical modeling approach, the results showed that WLP-AME yields improved formant frequencies for high-pitched sounds in comparison to the previously known methods (e.g., relative error in the first formant of the vowel [a] decreased from 11% to 3% when conventional LP was replaced with WLP-AME). Experiments conducted on natural vowels indicate that the formants detected by WLP-AME changed in a more regular manner between repetitions of different pitch than those computed by conventional LP.

Comparing measurement errors for formants in synthetic and natural vowels

“…Third, WLP-AME (Alku et al, 2013) was used with an order of 14, matching that of the Burg method. The algorithm depends on determining the glottal closure instant, using either the EGG signal (as reported in Alku et al, 2013) or the speech signal (as in the code we obtained from Manu Airaksinen).…”

“…The algorithm depends on determining the glottal closure instant, using either the EGG signal (as reported in Alku et al, 2013) or the speech signal (as in the code we obtained from Manu Airaksinen). For this study, we computed the glottal closure instants from the speech signal in order to create a fair comparison between synthetic and natural speech.…”

“…By contrast, the RS values for F1 had errors ranging from 0 to 8 Hz, and did not vary with F0. Alku et al (2013) used synthetic and natural vowels with F0 ranging from 100 to 450 Hz to test their weighted linear prediction with attenuated main excitation (WLP-AME) algorithm against five other linear prediction (LP) algorithms, all designed to measure the formants of high-F0 voices more accurately. The algorithm uses a temporal weighting function that attenuates the prediction error during the closed phase so that the times at which the residual error increases would not unduly warp the results of the analysis.…”

“…In the first, synthetic vowels were constructed with five different first formant (F1) values and nine different F0 values; formant bandwidths, and higher formant frequencies, were constant. Input formant values were compared to manual measurements and automatic measures using the linear prediction coding-Burg algorithm, linear prediction closed-phase covariance, the weighted linear prediction-attenuated main excitation (WLP-AME) algorithm [Alku, Pohjalainen, Vainio, Laukkanen, and Story (2013). J. Acoust.…”

A method for estimating vocal-tract shape from a target speech spectrum

“…The target spectrum was represented using cepstral parameters which were obtained through a linear predictive analysis. Therefore, using a pitch-robust speech analysis method such as discrete all-pole modeling [26] or the weighted linear prediction method [27] could improve the estimation accuracy for high-pitched speech signals.…”