Evaluation of orthogonal polynomial compression

In orthogonal polynomial compression, the short-term speech spectrum is first approximated by a family of orthogonal polynomials. The coefficients of each polynomial, which vary over time, are then adjusted in terms of their average value and range of variation. These adjustments can be used to compress (or expand) temporal variations in the average level, slope, and various forms of curvature of the short-term speech spectrum. The analysis and reconstruction of the short-term speech spectrum using orthogonal polynomials was implemented using a digital master hearing aid. This method of compression was evaluated on eight sensorineurally hearing-impaired listeners. Speech recognition scores were obtained for a range of compression conditions and input levels with and without frequency shaping. The results showed significant advantages over conventional linear amplification when temporal variations in the average level of the short-term spectrum were compressed, a result comparable to that obtained with conventional amplitude compression. A subset of the subjects showed further improvement when temporal variations in spectrum slope were compressed, but these subjects also showed similar improvements when frequency shaping was combined with level-only compression. None of the subjects showed improved speech recognition scores when variations in quadratic curvature were compressed in addition to level and slope compression.