Chiwei Che
Qiguang Lin
CAIP Ctr., Rutgers Univ., Piscataway, NJ 08855-1390
Cepstrum coefficients (including their different variants and/or derivatives) are nowadays commonly utilized in automatic speech recognition systems. Most adaptation algorithms for robust speech recognition also make use of cepstrum coefficients. It is therefore important to have a tool which can efficiently calculate and display cepstra. A computer program has been recently developed to display the cepstrogram of speech signals. The program has the following features: (1) It computes both linear prediction based and FFT based cepstra; (2) it displays cepstrum coefficients and their derivatives as a function of time and cepstrum index, as well as cepstrum coefficients at a fixed frame (section displays); (3) it generates synthesized spectra from cepstrum coefficients. In addition, the program can be used to segment and playback speech waveforms. The program has been utilized to study variations in cepstrum coefficients caused by speaker variability and channel differences. The tool is a useful supplement to conventional spectrograms for speech analysis. (The program is coded in C and runs in the X-window environment. It is available upon request.)