First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

A Pitch Detector Based on Event Detection Using the Dyadic Wavelet Tranform

Shubha Kadambe, Gloria F. Boudreaux-Bartels

Dept. of Elect. Engg., University of Rhode Island, Kingston, RI, USA

Several pitch detectors[l-9], which have been developed so far, are not always suitable for both low pitched and high pitched speakers and are not robust' to noise. In this paper, we describe an event based pitch detector which overcomes the above mentioned problems. We estimate the pitch period by detecting the glottal closure, which we label here as an event, using a time-scale representation viz., the Discrete Wavelet Transform(DWT) and by measuring the time interval between two such events. We illustrate the applicability of this pitch detector to a wide range of pitch periods i.e. for both low pitched and high pitched speakers and its robustness to noise with various examples. We then highlight the merits and demerits of this method by comparing it with existing pitch detection methods. The results of this study indicate that the algorithm works well for noise free and noisy vowels as well as voiced consonants.

Full Paper

Bibliographic reference.  Kadambe, Shubha / Boudreaux-Bartels, Gloria F. (1990): "A pitch detector based on event detection using the dyadic wavelet tranform", In ICSLP-1990, 469-472.