A nonlinear spectrum compression algorithm for the hearing impaired

A technique is proposed for increasing the speech intelligibility of persons with profound sensorineural hearing loss. The processing is based on a modification of the chirp-z transform that allows for nonlinear sampling along an arbitrary contour of the z-plane. Although the theory allows the nonlinear sampling to be arbitrary, an example showing that it is appropriate for nonlinear spectrum compression to aid the hearing-impaired is given. Samples are linearly spaced along the unit circle for a specified low-frequency segment; beyond the break-point the samples are nonlinearly spaced, perhaps spiraling inward to enhance spectral peaks, with the spacing width increasing with increasing frequency following an exponential function. A standard inverse DFT of these frequency samples returns the spectrum-compressed sequence. Processing of continuous speech requires windowing and processing of data in a manner similar to the short-time Fourier transform (STFT). The authors present the theoretical basis for, and development of, the algorithm