Intelligibility enhancement of noise-corrupted speech based on formant tracking involving prefiltering

The authors describe a technique for digitally processing a noise-corrupted speech signal in a manner designed to enhance its intelligibility. The technique is based on exacting speech features important for intelligibility and using these features to adaptively control algorithms that process the corrupted speech signal. The proposed technique is based on a novel processing algorithm for the estimation of formant frequencies. This technique involves prefiltering to emphasize each of the formants prior to applying linear prediction. Quantitative tests and preliminary listening tests indicate that the novel formant tracking technique offers some promise of improving the intelligibility enhancement capabilities over earlier methods