Recognition of spectrally degraded speech in noise with nonlinear amplitude mapping

The present study measured phoneme recognition as a function of the signal-to-noise level under conditions of spectral smearing and nonlinear amplitude mapping. Speech sounds were divided into 16 analysis bands. The envelope was extracted from each band by half-wave rectification and low-pass filtering and was then distorted by a power-law transformation whose exponents varied from a strongly compressive (p=0.3) to a strongly expanded value (p=3.0). This distorted envelope was used to modulate a noise which was spectrally limited by the same analysis filters. Results showed that phoneme recognition scores in a quiet condition were reduced only slightly with either expanded or compressed amplitude mapping. As the level of background noise was increased, performance deteriorated more rapidly for both compressed and linear mapping than for the expanded mapping. These results indicate that, although an expansive amplitude mapping may slightly reduce performance in quiet conditions, it may be beneficial in noisy listening conditions