Spectral analysis of speech signals using chirp group delay

• Authors: Baris Bozkurt;Thierry Dutoit;Laurent Couvreur
• Venue: Progress in nonlinear speech processing
• Year: 2007

This study presents chirp group delay processing techniques for spectral analysis of speech signals. It is known that group delay processing is potentially very useful for spectral analysis of speech signals. However, it is also well known that group delay processing is difficult due to large spikes that mask the formant structure. In this chapter, we first discuss the sources of spikes on group delay functions, namely the zeros closely located to the unit circle. We then propose processing of chirp group delay functions, i.e. group delay functions computed on a circle other than the unit circle in z-plane. Chirp group delay functions can be guaranteed to be spike-free if zero locations can be controlled. The technique we use here for that is to compute the zero-phased version of the signal for which the zeros appear very close (or on) the unit circle. The final representation obtained is named as the chirp group delay of zero-phased version of a signal (CGDZP). We demonstrate use of CGDZP in two applications: formant tracking and feature extraction for automatic speech recognition (ASR). We show that high quality formant tracking can be performed by simply picking peaks on CGDZP and CGDZP is potentially useful for improving ASR performance.