| This thesis is intended to investigate a new strategy for interpreting the human auditory physiology and assess the impact this knowledge might have on engineering applications such as speech encoding, word recognition, and speaker verification. The new interpretation is based around the signal processing properties associated with the autocorrelation function and has been developed from a strong foundation of literature that has been compiled over the last 50 years.; A distilation of the pertinent literature includes the phylogenetic development of audition, a detailed description of the anatomy of the inner ear, and the well-recognized inconsistencies involved with traditional interpretations of physiology and psychoacoustic experiments. These issues are discussed and reinterpreted to support a new philosophy on the signal processing functionality of the inner ear. From this conceptual model, a mathematical description and a partly phenomenological computer model are developed. The responses of the new model to test stimuli show qualitative similarity to the results of important psychoacoustic experiments, demonstrating that the new philosophy represents a more unified theory on audition than any proposed so far. In particular the new model predicts the effects known as "pitch", "the second filter", and some other phenomena that are not specifically recognized in the literature.; The knowledge gained in this effort is applied to techniques for speech encoding by real zeros, specifically, the reconstruction of signals that have been infinitely clipped. Finally, the concept is extended to include speculation on neural processes at the level of the cochlear nucleus in the brain. At this level an investigation is made to explore the power with which the new model can characterize phoneme signals generated by several different speakers. The results are shown to possess the historically accepted techniques for phoneme characterization as a subset of a dimensionally larger but stable description space. |
