A New Technique for the Estimation of Jitter and Shimmer of Voiced Speech Signal

This paper presents a new technique for the estimation of jitter and shimmer exploiting the detected value of the pitch-period and amplitude of voiced speech signal degraded by white noise. Perturbations of the glottis cycle duration (jitter) and intonation (shimmer) are, respectively, the irregularities of pitch-period and energy. We propose a modified circular autocorrelation function (MCACF) of the normalized hamming-windowed pre-filtered speech. A weighted and pitch-harmonic MCACF is utilized to effectively extract pitch-period for jitter computation. To carry out shimmer analysis, peak-to-peak amplitude of signal around the estimated pitch-period is optimally obtained by matching the windowed pre-filtered speech frame with a shifted impulse train. For performance evaluation vowels, such as "a, e, i, o, u", are used from the TIMIT database. The conventional algorithms manifest not only the distortion of pitch contours but also provides elevated values of the jitter and shimmer that can be interpreted as the presence of anomalies of the vocal tract, the glottis or the neuro-mechanism. On the contrary, for both clean and noisy speech, the proposed method results in a comparatively smooth pitch contour as well as small values of the jitter and shimmer confirming the presence of a normal voice with good stability and periodicity