Research Seminar - March19, 1999

Abstract

One of the fundamental tasks carried out by low-level auditory systems (of either a biological or an electronic nature) is the separation of an incoming signal into its multiple components.

Filtering the sound by cochlear or digital means results in parallel streams of information concerning a narrow frequency range, consisting of an amplitude (`loudness') and a phase component.

Much of the feature detection traditionally performed on the amplitude information can also be gleaned from the phase information. These features include the onset of a tone, its instantaneous frequency, the rate and direction of any frequency modulation, and more.

I will describe my recent attempts at phase-based feature detection in audio and some the problems encountered. Whilst not providing a panacea for some of the problems encountered with using amplitude information alone, phase can produce extra evidence maps to add to a pool of hypotheses as to what is actually going on in the sonic world.