Persons suffering from perceptive deafness commonly find it difficult to understand amplified speech, and their understanding of speech is easily destroyed by competing speech or noise. One source of perceptive distortion is recruitment, which exaggerates loudness differences among the acoustical elements of speech. A transposition of these distorted loudness relationships from the deaf-subject span of hearing to the normal span is illustrated graphically, and achieved in practice with an electronic processor. A recording of processed speech, simulating for normal listeners the loudness relationships perceived by deaf subjects with recruitment, accompanies this paper. Recruitment-compensation processing for hearing aids is also simulated. The recruitment simulation is validated by an experiment with four unilaterally deaf subjects, who compared processed speech in the normal ear with unprocessed speech in the impaired ear. The simulation suggests that (1) recruitment is a sufficient cause for loss of intelligibility in the deaf, whether or not there are other causes; (2) compensation for this recruitment is a necessary, although possibly insufficient, condition for restoring that intelligibility; (3) the benefit of using both compression and post-compression equalization in a hearing aid designed to compensate recruitment is likely to be considerably greater than the arithmetic sum of the separate, limited benefits of each process; and (4) the combined processing, by restoring redundant speech-recognition cues to the subjects's perception, can increase the resistance of intelligibility to acoustical interference.

A deaf person with recruitment perceives sound as though listening through a volume expander followed by an attenuator, the expansion ratio and attenuation be'rag typically frequency dependent. (Other perc•tive aberrations may also be present, of course.) The subject is often prevented from using enough hearing-aid gain to bring weak consonants into the useful dynamic range of his hearing, because this amount of gain would make lower-frequency, high-amplitude vowels intolerably loud. Such subjects commonly find amplified speech to have poor intelligibility. In a preliminary experiment it is established that recruitment in normal subjects, induced by masking or simulated by expansion of the signal, reduces the intelligibility of amplified speech severely, and that this intelligibility can be largely restored by signal processing. The implication is that recruitment in deaf subjects is a sufficient cause for loss of intelligibility, whether or not there are other causes. In the pre•ent experiments, speech is processed by a two-channel amplitude compressor whose frequency-dependent compression ratio is adjusted to compensate the recruitment of the individual subject, and the compressed speech i• subjected to frequency-selective amplification similarly adapted to the subject. The aim is to amplify each acoustical element of speech, at each frequency-amplitude coordinate of the speech band, to a relative 1oudne• for the deaf subject corresponding to the relative loudness of that speech element perceived by normals. Thi• processing improved speech recognition, both in quiet and in the presence of competing speech introduced before processing, for six perceptively deaf subjects. Subjects showed an improvement in either initial-or terminal-consonant recognition of at least 22% and as much as 160% at optimum levels in quiet, and from 10% to 229% with speech interference 10 dB below the pre-processed signal.The compressor characteristics that can be controlled by the designer include the attack and release times, the compression ratio (the ratio, expressed arithmeti-cally, of input-level change in decibels to output-level change), the compression linearity (constancy of the compression ratio over the dynamic range), and the compression threshold (the input level at which gain begins to decrease with input). A. Attack and Release TimesThe reduction of compressor gain in response to an increase of signal voltage--the compression attack-can be accomplished in a small fraction of 1 msec without creating audible distortion. The release of compression, however, must be slowed up to prevent compressor action from following the instantaneous amplitude of individual cycles, an operating mode that would introduce severe waveform distortion.Commercial compressors rarely provide a release time 2 shorter than several hundred milliseconds. A delay of this length following a high-amplitude vowel would keep the compressor in its low-gain state for the duration of a succeeding consonant, leaving the vowelconsonant amplitude ratio unchanged...

Electronic models that process signals to simulate omissions and distortions in perception by the deaf are useful analytical tools. These models can help in determining the degree to which the represented loss and/or distortion of acoustic speech cues, isolated from' any central pathology, affects speech intelligibility. They can also help in evaluating the potential effectiveness of different modes of compensatory signal processing. Three examples of such models, designed on the basis of measured characteristics of the deaf subjects' residual hearing, are described. A model of recruitment combined with accentuated high-frequency loss is used to study speech perception in noise. It suggests an explanation for the previously reported ability of sensorineurals to tolerate the masking of speech by white noise at least as well as normals do, and for the special vulnerability of the speech perception of such deaf subjects to masking signals that have the spectral distribution of speech. A second model, that simulates the severely restricted dynamic range of hearing in profound deafness, predicts that this one factor can destroy the intelligibilty of amplified speech by making the perceived sound drop out at frequent intervals. Finally, a model that simulates reduced frequency-discriminating ability is described, as an example of a model that deals with aspects of perception other than loudness.

The range of combined intersubject and cushion-fit variation in the response of the TDH-39 audiometer earphone in an MX41/AR cushion, reported in the literature and confirmed in these measurements, is of the order of 25 dB at 100 Hz, 6 dB at 1 kHz, and 15 dB at 5 kHz. The performance of this earphone system is further impaired by cushion-induced physiological noise that masks low-frequency thresholds, typically by 6 dB in normal-hearing subjects. An experimental mounting for the TDH-39 is described and analyzed; it is a circumaural device designed to keep the earphone-ear acoustical system as near constant as possible with successive fittings and different subjects, and to reduce physiological noise. The performance of the TDH-39 in this mounting, compared to performance in an MX41/AR cushion, showed a substantial reduction in cushion-fit and intersubject response variation, and masking by physiological noise was largely eliminated in the subjects tested. The mounting has an optional cup for reducing the transmission of external noise, but this reduction is at the expense of a partial return of physiological noise masking. Average midrange sensitivity was reduced 5–6 dB. Response measurements were made with a free-field reference method, using a probe microphone at the subject's eardrum.

According to Plomp [J. Acoust. Soc. Am. 83, 2322–2327 (1988)], fast multichannel amplitude compression has a predictably negative effect on speech intelligibility for both normal and impaired listeners. The following letter is a response to that conclusion.

Hearing aid design to alleviate the noise problem has concentrated on improving the signal-to-noise ratio with the aid, using devices such as directional microphones, adaptive filters, and circuits that discriminate between steady-state noise and speech. The design approach discussed here is directed at improving the speech recognition of hearing-impaired listeners at a given signal-to-noise ratio, by restoring to their perception speech cues they no longer hear because of their impairment. This allows them to retain more of the redundant information in speech after masking has taken its toll, and empowers their ability to separate desired from undesired signals (what Broadbent calls "selective listening" in persons with normal hearing). Experimental results are presented.

A method of making in-the-ear probe-microphone measurements is described, using readily available equipment. The procedure is designed to make such measurements easier and safer than they have been in the past.Subject Classification: 85.62; 65.80.

An electronic model simulating the severely restricted dynamic range of hearing typical in profound deafness has been used to investigate the effect of this restriction on speech perception, and also to investigate the potential effect of signal processing designed to improve intelligibility under such conditions. The model is patterned on the average measured thresholds and discomfort levels of a group of profoundly deaf children. It rejects signal elements whose levels fall below electronic thresholds in each of ten adjacent 13-octave bands, and amplitude expanders in each band simulate the accelerated loudness growth of recruitment. Speech passed through the model is broken up by periods of silence that occur whenever signal levels drop below thesholds in all bands, and the speech becomes unintelligible. Three-channel amplitude compression restores the rhythmic pattern and some intelligibility to the speech. When compression is combined with synthetic fricatives (placed at 800-Hz center frequency), intelligibility is increased further. The performance of this model suggests that previous attempts to improve speech reception for the profoundly deaf by signal processing may have been compromised by a failure to deal with the problem of a very narrow dynamic range of hearing.