Feature-map methods for extracting sound frequency modulation

The role of frequency modulation (FM) in the auditory system is investigated. Psychoacoustic evidence shows that FM is a cue for noncontextual auditory scene analysis, and neurophysiological evidence shows that it is detected by neurons in some animal auditory systems. Some filters for detecting FM in sound are presented. One method uses correlation of a kernel with a time-frequency representation of cochlear neural impulses of a sound to reveal its FM components. Another correlation method is based on detecting FM in channel-by-channel autocorrelation images, with some data reduction and sharpening applied. The filters are tuned for various spatiotemporal frequencies to extract information about sound with various FM characteristics, producing maps showing time-frequency points at which the features are present. These maps can allow another process to decide what sources are present or to use FM cues for pattern recognition