Evaluation of a technique involving processing with feature extraction to enhance the intelligibility of noise-corrupted speech

Results obtained in an experimental evaluation of speech intelligibility of an adaptive processing technique designed to enhance the intelligibility of speech in noise are presented. The technique relies on speech features extracted from noise-corrupted speech to control a cepstral processor. The speech-processing algorithm has been improved over the fundamental technique described by R.J. Conway and R.J. Niederjohn (1987). The evaluation method makes use of a computer implementation of a modified version of the diagnostic rhyme test. Results for four subject and several signal-to-noise ratios are presented.<>