Harmonic Tracking-based Short-Time Chirp Analysis of Speech Signals

The Short-Time Fourier Transform is the most popular timefrequency analysis tool applied in speech processing. This transform delivers fair quality analysis for periodic signals, but, since speech is quasi-periodic, the transform suffers from blurry harmonic representation when voiced speech undergoes changes in pitch. This frequency variation could be relative high in comparison with the analysis window length, this leading to inconsistent bins in frequency or time domain. This paper is focused on a method to achieve accurate time-frequency representation of speech signals also with changing pitch, and is based on a new self-adaptive analysis tool, called the Short-time Chirp transform (STChT). The basis of this transform is composed of quadratic chirps whose chirp-rate is updated for each analysis segment according to the pitch evolution. The pitch trajectory is estimated segment-by-segment by a new harmonicity analysis tool introduced in this paper. The Short-Time Chirp Transform offers more detailed time-frequency representation of speech signals especially with changing pitch as occurs in normal intonation.