Simulation of increased masking in sensorineural hearing loss for a preliminary evaluation of speech processing schemes

Sensorineural loss is characterized by increased hearing threshold, reduction in the dynamic range of hearing and loudness recruitment, and increased temporal and spectral masking, resulting in degraded speech perception. Several techniques including spectral contrast enhancement, multi-band frequency compression, and dichotic binaural presentation have been investigated for reducing the adverse effects of increased masking. Assessment of speech processing techniques and optimization of processing parameters involves listening tests on hearing-impaired listeners. These tests are time consuming and may cause a fatigue, particularly in elderly subjects. A simulation of hearing loss, by processing the speech signal through a model of the loss characteristics, is useful in conducting the listening tests on normal-hearing subjects, for a preliminary evaluation of the schemes and particularly for selecting the processing parameters. The present study used addition of broad-band noise, band-limited to speech frequency range, at a specific SNR with respect to short-time (10 ms) energy of the signal. Different levels of loss were simulated by varying the SNR. In this simulation, no noise gets added during silence segments. Listening tests to assess the loss simulation were conducted using three types of test material: vowel-consonant-vowel (VCV) utterances, phonetically balanced word lists, and modified rhyme test. Recognition score from subject responses was used as a measure of speech intelligibility and response time was used as a measure of load on the perception process. For all the three test materials, decrease in the recognition scores and increase in response times for normalhearing subjects showed the same pattern as the corresponding results for subjects with moderate-to-severe sensorineural loss. A relative information transmission analysis of the stimulus-response confusion matrices for VCV utterances showed that the simulated loss did not affect reception of voicing and nasality features and it had maximum adverse effect on the reception of place and duration features, indicating that the addition of broadband noise with constant SNR with respect to short-time signal energy simulated an increased spectral and temporal masking.