TIME-SCALING OF CONSONANT-VOWEL TRANSITIONS USING HARMONIC-PLUS-NOISE MODEL FOR IMPROVING SPEECH PERCEPTION BY LISTENERS WITH MODERATE SENSORINEURAL IMPAIRMENT PACS: 43.71.Ky

Speech produced to improve intelligibility in a difficult communication environment or for hearing impaired listeners is called clear speech. It is generally characterized by consonant segments with increased intensity and expanded formant transition duration, burst duration, and voice onset time. These characteristics are likely to reduce the effect of increased masking associated with sensorineural impairment, and may improve speech intelligibility. Our objective is to develop a technique for automated conversion of conversational speech to clear speech, by time-scale modification and intensity enhancement of consonant segments, using an analysissynthesis platform based on harmonic plus noise model (HNM). The transition segments are automatically located by a landmark detection algorithm. Consonant-vowel and vowelconsonant transition segments are time-expanded and the steady state vowel segments are appropriately compressed to maintain the original speaking rate. Clearly intelligible speech could be produced for time-scaling factors in the range 1 to 2. The method was evaluated using VCV syllables on normal hearing subjects with simulated hearing loss. This technique can also be combined with other speech processing techniques for hearing aids, like multi-band compression, spectral modification, and binaural dichotic presentation