Correction of the peripheral spatiotemporal response pattern: a potential new signal-processing strategy

Perception of across-frequency asynchrony by listeners with cochlear hearing loss

Cochlear hearing loss is often associated with broader tuning of the cochlear filters. Cochlear response latencies are dependent on the filter bandwidths, so hearing loss may affect the relationship between latencies across different characteristic frequencies. This prediction was tested by investigating the perception of synchrony between two tones exciting different regions of the cochlea in listeners with hearing loss. Subjective judgments of synchrony were compared with thresholds for asynchrony discrimination in a three-alternative forced-choice task. In contrast to earlier data from normal-hearing (NH) listeners, the synchronous-response functions obtained from the hearing-impaired (HI) listeners differed in patterns of symmetry and often had a very low peak (i.e., maximum proportion of "synchronous" responses). Also in contrast to data from NH listeners, the quantitative and qualitative correspondence between the data from the subjective and the forced-choice tasks was often poor. The results do not provide strong evidence for the influence of changes in cochlear mechanics on the perception of synchrony in HI listeners, and it remains possible that age, independent of hearing loss, plays an important role in temporal synchrony and asynchrony perception.

Noise-induced hearing loss alters the temporal dynamics of auditory-nerve responses

Auditory-nerve fibers demonstrate dynamic response properties in that they adapt to rapid changes in sound level, both at the onset and offset of a sound. These dynamic response properties affect temporal coding of stimulus modulations that are perceptually relevant for many sounds such as speech and music. Temporal dynamics have been well characterized in auditory-nerve fibers from normal-hearing animals, but little is known about the effects of sensorineural hearing loss on these dynamics. This study examined the effects of noise-induced hearing loss on the temporal dynamics in auditory-nerve fiber responses from anesthetized chinchillas. Post-stimulus-time histograms were computed from responses to 50-ms tones presented at characteristic frequency and 30 dB above fiber threshold. Several response metrics related to temporal dynamics were computed from post-stimulus-time histograms and were compared between normal-hearing and noise-exposed animals. Results indicate that noise-exposed auditory-nerve fibers show significantly reduced response latency, increased onset response and percent adaptation, faster adaptation after onset, and slower recovery after offset. The decrease in response latency only occurred in noise-exposed fibers with significantly reduced frequency selectivity. These changes in temporal dynamics have important implications for temporal envelope coding in hearing-impaired ears, as well as for the design of dynamic compression algorithms for hearing aids.

Perception of temporally processed speech by listeners with hearing impairment

OBJECTIVES

The purpose of this study was to assess a new signal-processing strategy, Spatiotemporal Pattern Correction (SPC), for a group of listeners with varying degrees of sensorineural hearing loss. SPC is based on a physiological model of the level-dependent temporal response properties of auditory nerve (AN) fibers in normal and impaired ears. SPC attempts to "correct" AN response patterns by introducing time-varying group delays that differ across frequency channels. Listeners' speech intelligibility and preference judgments were obtained at different strengths of SPC processing.

DESIGN

Fifteen listeners, 12 with bilateral sensorineural hearing loss and 3 with normal hearing, participated in this study. Listeners with hearing impairment were separated into 3 groups (mild, moderate, and moderate-to-severe), based on their pure-tone averages at 0.5, 1, 2, and 4 kHz. All preference judgments were made using sentences from the Hearing-In-Noise-Test (HINT). The sentences were processed at five SPC strengths (1.0, 1.1, 1.2, 1.3, and 1.4). At an SPC strength of 1.0, the speech was processed through the SPC signal processor, becoming bandlimited, but no dynamic group delays were added to the signal. On each trial, listeners were presented a single sentence at two different SPC strengths, 1.0 (uncorrected) and a randomly selected SPC strength. Listeners were instructed to identify which of the two sentences they preferred. Sixteen vowel-consonant syllables, a subset of the Nonsense Syllable Test (NST), were used to assess speech intelligibility at three SPC strengths (1.0, 1.1, and 1.2). Listeners were asked to push one of 16 buttons on a response box that corresponded to the vowel-consonant syllable they heard. The moderate-to-severe hearing loss group was also tested using the low-probability sentences of the Speech-Perception-In-Noise (SPIN) test.

RESULTS

Listeners with normal hearing and listeners with mild hearing loss preferred the quality of uncorrected sentences (SPC strength of 1.0) compared with SPC-processed sentences. As the strength of SPC increased, listeners' preference for uncorrected sentences also increased. Listeners with moderate and moderate-to-severe hearing loss had difficulty differentiating between uncorrected and lower strength SPC-processed sentences but strongly preferred the uncorrected sentences as compared with high strength SPC-processed sentences. Vowel-consonant intelligibility was not improved or degraded by SPC. Sentence recognition scores for listeners with moderate-to-severe hearing loss decreased with SPC processing.

CONCLUSIONS

Although SPC-processed speech was not preferred by listeners with hearing loss, the listeners with a more moderate degree of impairment could not differentiate the unprocessed sentences from the SPC processed sentences. Speech intelligibility was not improved by SPC processing.

Estimates of basilar-membrane nonlinearity effects on masking of tones and speech

OBJECTIVE

The aim of this experiment was to assess the contribution of cochlear nonlinearities to speech recognition in noise for individuals with normal hearing and a range of quiet thresholds. For signals close to the characteristic frequency (CF) of a place on the basilar membrane, the normal growth of response of the basilar membrane is linear at lower signal levels and compressed at medium to higher signal levels. In contrast, at moderate to high CFs, the basilar membrane responds more linearly to stimuli at frequencies well below the CF regardless of input level. Thus, for moderate-level speech and a lower frequency masker, the response to the masker grows linearly whereas the response to the speech is compressed, which may result in changes in the effectiveness of the masker on speech recognition with increases in masker level. To test this hypothesis, observed speech-recognition scores were compared with scores predicted using an audibility-based model, which did not include nonlinear effects that may influence masker effectiveness.

DESIGN

Growth of simultaneous masking was measured for moderate-level bandpass-filtered nonsense syllables and for 350-msec pure tones at frequencies within the speech passband. Masker frequencies were within (on-frequency) or below (off-frequency) the speech passband. Estimates of basilar-membrane nonlinearities were derived from growth-of-masking functions for 10-msec, 2.0- and 4.0-kHz tones in narrowband, off-frequency maskers presented simultaneously. Subjects were 26 adults with normal hearing with approximately a 20-dB range of average quiet thresholds.

RESULTS

Breakpoints (i.e., the levels corresponding to the transitions from linear to nonlinear responses) were strongly associated with quiet thresholds but slopes measured above the breakpoints were independent of quiet thresholds. Individual differences were substantially larger for off-frequency masking of pure tones and speech than for on-frequency masking of pure tones and speech. Using an audibility-based predictive model, the change in speech audibility resulting from the compressed response to speech with increasing off-frequency masker level (and the resulting decline in scores) was well predicted from nonlinear growth of masking for pure tones measured in the same off-frequency masker. However, absolute speech-recognition predictions were generally inaccurate and were a function of how well pure-tone signal levels at masked threshold estimated masker effectiveness for speech. That is, subjects with lower off-frequency masked thresholds had less accurate predictions of speech recognition in off-frequency maskers.

CONCLUSIONS

Large individual differences in off-frequency masking of pure tones and speech are consistent with the assumption that small changes in the shape of the basilar-membrane input-output function result in large changes in the amount of off-frequency masking but small (if any) changes in on-frequency masking where the signal and masker are subject to a similar compression. Growth of off-frequency masking of pure tones and speech were correlated with each other, consistent with the underlying basilar-membrane response, and consistent with changes in breakpoints for subjects with normal hearing and a range of quiet thresholds. These results provide support for a role of nonlinear effects in the understanding of speech in noise.