Nonlinear Cochlear Signal Processing and Masking in Speech Perception

Auditory masking is critical to our understanding of speech and music processing. There are many classes of masking, but two major classes are easily defined. These two types of masking and their relation to nonlinear (NL) speech processing and coding are the focus of this chapter. The first class of masking, denoted neural masking, is due to internal neural noise, characterized in terms of the intensity just noticeable difference, denoted ∆I(I, f, T ) (abbreviated JNDI ) and defined as the “just discriminable change in intensity.” The JNDI is a function of intensity I , frequency f and stimulus type T (e.g., noise, tones, speech, music, etc). As an internal noise, the JNDI may be modeled in terms of a loudness (i.e., perceptual intensity) noise density along the length of the cochlea1 (0 ≤ X ≤ L), described in terms of a partial loudness JND (∆L(X,T ), a.k.a. JNDL). The loudness JND is a function of the partial loudness L(X), defined as the loudness contribution coming from each cochlear critical band, or more generally, along some tonotopic central auditory representation. The critical band is a measure of cochlear bandwidth at a given cochlear place X . The loudness JND plays a major role in speech and music coding since coding quantization noise may be masked by this internal quantization (i.e., “loudness noise”). The second masking class, denoted here as dynamic-masking, comes from the NL mechanical action of cochlear outer hair cell (OHC) signal processing. It can have two forms, simultaneous and