Orthogonal polynomial compression amplification for the hearing impaired
Harry N. Levitt, Arlene C. Neuman, Jayashree Toraskar · The Journal of the Acoustical Society of America · 1986
The method of orthogonal polynomials is a computationally concise curve-fitting technique [Levitt and Rabiner, J. Acoust. Soc. Am. 49, 569–582 (1971)]. Each coefficient in the polynomial expansion is obtained using a single vector operation. The computational simplicity of the procedure allows for real-time speech processing using a digital master hearing aid configured around a MAP-300 array processor. The short-term spectrum of speech is first approximated by a set of orthogonal polynomials and then modified by compressing the range of variation of the polynomial coefficients. Compression of the constant term is analogous to conventional amplitude compression. Compression of the linear coefficient reduces variations in the slope of the short-term spectrum. Compression of the quadratic coefficient reduces variations in convex/concave curvature. Implementation in real time allowed for a paired-comparison adaptive strategy to be used in searching for the optimum set of compression coefficients for each subject. Compression of the constant term was found to have the largest effect leading to a significant improvement in the dynamic range of the performance-intensity function for a group of hearing impaired subjects. [Research supported by NINCDS.]