Use of periodicity and jitter as speech recognition features

Abstract

addition of new feature components to the frame vectors.We investigate a class of features related to voicing parameters that indicate whether the vocal chords are vibrating. Features describing voicing characteristics of speech signals are integrated with an existing 38-dimensional feature vector consisting of first and second order time derivatives of the frame energy and of the cepstral coefficients with their first and second derivatives. HMM-based connected digit recognition experiments comparing the traditional and extended feature sets show that voicing features and spectral information are complementary and that improved speech recognition performance is obtained by combining the two sources of information. 21 0-7803-4428-6/98 $10.00 0 I998 IEEE

Robust excitation-based features for Automatic Speech Recognition

“…Another common approach to quantify periodicity relies on the auto-correlation (AC) function of the speech signal [9,11] by measuring the relative height of the maximum of this function in the plausible pitch range. The Average Magnitude Difference Function (AMDF) can be formulated as a function of the AC function.…”

“…The first attempt was made by Thomson [9,10] who proposed the use of two voicing measures: an auto-correlation based measure of periodicity and the jitter to characterize the inter-frame pitch variation. When combined to cepstral features, a relative reduction of 40% of the string error rate was obtained on a connected digit recogntion task.…”

Acoustic Feature Combination for Robust Speech Recognition

“…There have been also attempts at using articulatory information in the acoustic front-end, e.g. autocorrelation based voicedness feature [1]. In this paper we investigate the combination of different auditory based and articulatory based acoustic features.…”

“…In [1], liftered cepstral coefficients derived from all-poles magnitude spectrum has been directly concatenated with a voicedness feature. Using the concatenated features, a large relative improvement in word error rate (WER) has been achieved by applying discriminative training.…”

Characterizing Feature Variability in Automatic Speech Recognition Systems

“…In [6], a sub-optimal solution is proposed for selecting features from two different sets. Other approaches integrate some specific parameters into a single stream of features [9]. Without attempting to find an optimal set of acoustic measurements, many recent automatic speech recognition (ASR) systems combine streams of different acoustic measurements ( [4], [10]).…”