Speech Analysis with the Short-Time Chirp Transform

摘要

The most popular time-frequency analysis tool, the Short-Time Fourier Transform, suffers from blurry harmonic representation when voiced speech undergoes changes in pitch. These relatively fast variations lead to inconsistent bins in frequency domain and cannot be accurately described by the Fourier analysis with high resolution both in time and frequency. In this paper a new analysis tool, called Short-Time Chirp Transform is presented, offering more precise time-frequency representation of speech signals. The base of this adaptive transform is composed of quadratic chirps that follow the pitch tendency segment-by-segment. Comparative results between the proposed STCT and popular time-frequency techniques reveal an improvement in time-frequency localization and finer spectral representation. Since the signal can be resynthesized from its STCT, the proposed method is also suitable for filtering purposes.