The purpose of this research is further the development and evaluation of auditory models that can enhance the performance of automatic speech recognition systems. The research has two specific objectives. The primary objective is to extend an auditory model developed by Weintraub (1986, 1985, 1984) to incorporate acoustic information sources (e.g. both periodicity and spectral information) that can be used to separate targeted speech from unwanted sounds, including other speech. The additional acoustic information should result in a significant improvement of performance when abstracting speech from a background of noise. A Second objective is to develop a new methodology for evaluating a range of signal-processing algorithms, including auditory modeling. This method is based on an information- theoretic approach that follows the ideal-observer concept. With this method, the information actually extracted by a signal- processing algorithm can be compared with the information potentially available in the signal, to obtain a measure of the performance for each algorithm. The project plans to implement several well-known signal processing algorithms for dealing with speech in noise. The result of the of this research allow us to evaluate modifications to the auditory model and compare the performance achieved to that of other signal processing algorithms.
