Representation of fine structure of speech signals using time-frequency zeros

In this paper, we propose a novel time-frequency analysis method for speech, which consists of 1) subband decomposition by the holomorphic filterbank proposed by us, and 2) detection of time-frequency zeros. By theoretical analysis and experiments, we show that 1) voiced speech generates a stable two-dimensional (2-D) maps of zeros which are different vowels by vowels, 2) it represents pitch structure, formant structure, and glottal excitation instance, which are important for analyzing voiced speech in frequency and time domain.