Signal processing strategies that attempt to mimic the frequency resolution of the healthy cochlea require finer frequency resolution at low frequencies than at high. The filtering required will cause more delay to the low-frequency end of the spectrum than to the high-frequency end, which may have disturbing effects. In a real-time application, making the delay constant across frequency would require the higher frequency components to be subjected to a compensating delay, leading to a larger overall delay, which may also give rise to disturbing effects (Stone & Moore, 1999, 2002). This experiment assessed the effects of frequency-dependent delay on subjective and objective measures of speech production and perception.
Digital filtering was used to introduce across-frequency delays of 0, 4, 9, 15, or 24 msec. The low frequencies were always more delayed than the high frequencies, with a smooth variation in delay between the low and the high frequencies. In addition, there was a 2.5 msec delay that did not vary with frequency arising from the processes of analog-digital-analog conversion and equalization. Ten subjects with symmetric, bilateral, moderate hearing impairment of cochlear origin were fitted binaurally with Phonak PicoNet2 BTE linear hearing aids. The aids were programmed to provide insertion gains closely approximating the NAL (RP) (Byrne & Dillon, 1986) prescription for each subject and ear. The output of an omnidirectional microphone, attached on top of each aid, fed a digital filtering unit that introduced the across-frequency delay. Audio shoes reintroduced the signal back into the BTE aids. Subjects were tested in a counterbalanced order for their identification of vowel-consonant-vowel (VCV) syllables. In a second session, the same subjects spoke from a script: their utterances were recorded on tape, speech production rates were measured and subjective ratings of the disturbance of the delay were obtained. Subjects required some training to recognize the effects of the delay and rate them consistently.
Subjective disturbance increased progressively with increasing across-frequency delay. A 9-msec delay was significantly more disturbing than a 0-msec delay. A delay of about 20 msec led to a mean rating of “disturbing.” VCV identification decreased significantly once the across-frequency delay was 15 msec or greater. However, word production rates were not significantly affected by across-frequency delay over the range tested.
Relatively small across-frequency delays (9 to 15 msec) give rise to significant changes in speech identification and subjective disturbance. A delay of 9 to 15 msec, constant across frequency, would have a smaller effect (Stone & Moore, 1999). It appears desirable to compensate, at least in part, for across-frequency delays introduced by any processing.