We used real-time processing in a wearable digital hearing aid to examine the effect of processing delay on normal-hearing participants while speaking. Objective and subjective data were recorded so as to permit analysis of both the production and perception of speech read aloud from a script. We also asked participants to rate the disturbance of the echo introduced by the delay.
Thirty-two (16M, 16F) participants were fitted binaurally with behind-the-ear (BTE) aids connected to a digital processor. A 4 mm Libby horn, surrounded by an expanding foam earplug, conducted processed sound into each ear canal. The processor provided either linear processing or three-channel, fast-acting wide dynamic range compression, independently to each ear. Insertion gains were set, using a KEMAR manikin, to be 0 dB over a wide frequency range, for frontally presented speech with a free field level of 65 dB SPL. Additionally, the aids introduced one of four selectable delays (7 to 43 msec) between the BTE microphone and receiver. After a short period of acclimatization, each participant read 16 prose passages of about 500 words in length in each of two similar-sized rooms with markedly different acoustics: reverberant and nonreverberant. For each passage, a subjective rating of the level of disturbance of the perceived echo was recorded, as well as simultaneous recordings from a microphone and a Laryngograph, which directly records glottal pulses.
Disturbance ratings generally increased monotonically with increasing delay. Averaged results show that a delay between 25 and 30 msec is rated as “disturbing.” Measures were also taken of word production rate, speech level and range of level as well as fundamental frequency and range of fundamental frequency. For these measures of speech production, there was no significant effect until the delay exceeded 30 msec. There was little effect of acoustic environment or aid processing (linear or compression).
The acceptability of delays introduced by digital hearing aids is primarily determined by aspects of the perception of self-generated speech. Speech production, on average, is hardly affected unless the processing delay exceeds 30 msec. The permissible limit of 20 to 30 msec is smaller than the delays at which audio-visual integration is disrupted.