Bayesian estimation of high-parameter models of the auditory filter

In this paper we describe a method for comparing frequency selectivity in simultaneous and forward masking. The method is designed to eliminate off-frequency listening, which may have had a confounding influence in earlier studies. Thresholds for 1-kHz sinusoidal signals were measured as a function of the width of a spectral notch, centered at 1 kHz, a noise masker. In experiment I thresholds were measured in forward masking for signal durations of 5, 15, and 45 ms, with a noise spectrum level of 40 dB SPL/Hz. Thresholds decreased with increasing notch width for all signal durations. However, the change was more gradual at longer signal durations. This is consistent with a model in which the process of decay of masking follows the auditory filter. For each signal duration, threshold was also measured as a function of the level of a noise without a spectral notch. Each signal threshold for the notch-noise condition was then expressed as the level of a flat-spectrum noise which would give the same masking. When transformed in this way, the data for the three signal durations coincide, and can be interpreted in terms of the same auditory filter. In experiment II the 5-ms signals, whose level was fixed, followed the masker with delays of 5, 15, or 25 ms. The noise spectrum level was varied to find threshold. The change in threshold with notch width was independent of signal delay, confirming that the process of decay of masking follows the auditory filter. Experiment III showed that short-term spectral changes produced by differences in the shapes of the envelopes of signal and masker did not influence the results. In experiment IV thresholds for a 5-ms signal were measured in simultaneous and forward masking, both for a fixed noise level and for a fixed signal level. The derived auditory filters are similar for the two types of measure for each type of masking. However, the auditory filters derived in forward masking have bandwidths 17% smaller and slopes 48% greater than those in simultaneous masking. The differences between simultaneous and forward masking are interpreted in terms of suppression.

A Bayesian adaptive procedure, the quick-auditory-filter (qAF) procedure, was used to estimate auditory-filter shapes that were asymmetric about their peaks. In three experiments, listeners who were naive to psychoacoustic experiments detected a fixed-level, pure-tone target presented with a spectrally notched noise masker. The qAF procedure adaptively manipulated the masker spectrum level and the position of the masker notch, which was optimized for the efficient estimation of the five parameters of an auditory-filter model. Experiment I demonstrated that the qAF procedure provided a convergent estimate of the auditory-filter shape at 2 kHz within 150 to 200 trials (approximately 15 min to complete) and, for a majority of listeners, excellent test-retest reliability. In experiment II, asymmetric auditory filters were estimated for target frequencies of 1 and 4 kHz and target levels of 30 and 50 dB sound pressure level. The estimated filter shapes were generally consistent with published norms, especially at the low target level. It is known that the auditory-filter estimates are narrower for forward masking than simultaneous masking due to peripheral suppression, a result replicated in experiment III using fewer than 200 qAF trials.

The behaviour of delayed evoked otoacoustic emission (DEOAE) has been studied in normally hearing adults under a conventional forward-masking paradigm, and subjective measurements were carried out additionally for comparison. The clicks served as signals and the noise bursts were used as masker. In different experimental sets, signal and masker intensity, masker duration, and the interval between masker and maskee were altered. At masker levels corresponding to the subjective post-masking threshold of the clicks, the DEOAE was unaffected, i.e. had no noticeable alteration, compared with click stimulation without masking. Even at higher masker levels the inaudible clicks elicited clearly discernible DEOAEs. The forward-masking detection threshold of DEOAE ('DEOAE post-masking threshold') was reached at masker levels approximately 35 dB above the subjective post-masking threshold. The gap between subjective and DEOAE post-masking threshold vis-à-vis the masker level was also evident at different masker durations and different time intervals between masker and maskee. Central neural and peripheral receptor mechanisms are suggested to be the constituents of the masking phenomenon. The neural mechanism is involved at low masker levels. The receptor mechanism effectively joins the neural one at masker levels exceeding the threshold of psychoacoustic masking. The progressive increase in the number of auditory units from the periphery to the centre in the hearing system, linked with an increase in inhibition, can help to explain these effects.

The frequency selectivity of an auditory filter system is often conceptualized as a bank of bandpass auditory filters. Over the past 30 years, many simultaneous masking experiments using notched-noise maskers have been done to define the shape of the auditory filters (e.g., Glasberg and Moore 1990; Patterson and Nimmo-Smith 1980; Rosen and Baker, 1994). The studies of Glasberg and Moore (2000) and Baker and Rosen (2006) are notable inasmuch as they measured the human auditory filter shape over most of the range of frequencies and levels encountered in everyday hearing. The advantage of using notched-noise masking is that one can avoid off-frequency listening and investigate filter asymmetry. However, the derived filter shapes are also affected by the effects of suppression. The tunings of auditory filters derived from data collected in forward masking experiments were apparently sharper than those derived from simultaneous masking experiments, especially when the signal levels are low. The tuning of a filter is commonly believed to be affected by cochlear nonlinearity such as the effect of suppression. In past studies, the tunings of auditory filters derived from simultaneous masking data were wider than those of filters derived from nonsimultaneous (forward) masking data (Moore and Glasberg 1978; Glasberg and Moore 1982; Oxenham and Shera 2003). Heinz et al. (2002) showed that a tuning is generally sharpest when stimuli are at low levels and that suppression may affect tuning estimates more at high characteristic frequencies (CFs) than at low CFs. If the suggestion of Heinz et al. (2002) holds, i.e., if suppression affects frequency changes, comparing the filter bandwidths derived from simultaneous and forward masking experiments would indicate this. In this study we attempt to estimate filter tunings using both simultaneous and forward masking experiments with a notched-noise masker to investigate how the effects of suppression affect estimates of frequency selectivity across signal frequencies, signal levels, notch conditions (symmetric and asymmetric), and signal delays. This study extends the study of Unoki and Tan (2005).

Growth of masking for OFF-frequency conditions (probe frequency above the masker spectrum) and ON-frequency conditions (probe within the masker spectrum) was investigated using simultaneous masking in three subjects with normal hearing and nine subjects with high-frequency sensorineural hearing loss. Growth-of-masking functions (probe thresholds as a function of masker intensity) for OFF-frequency conditions were obtained for probe tones placed at six frequencies above a 200-Hz-wide masker with an upper edge at 520 Hz. Growth-of-masking functions for ON-frequency conditions were obtained for probe tones placed within the 200-Hz-wide masker and for probe tones placed within 400-Hz-wide maskers with upper edges at 1040, 1300, 1627, and 2040 Hz (probe tones placed 20 Hz below the upper edge frequency). Growth-of-masking functions were fit with a power function of masker intensity added to an internal noise with intensity equal to the absolute threshold for the probe, and were well described by two free parameters and a threshold constant: the growth-of-masking slope (beta), a masking sensitivity constant (kappa) that indicated the minimum effective masker level at which masking began, and the intensity of the probe at absolute threshold (IT). For OFF-frequency masking conditions, growth-of-masking slopes (beta) decreased by a factor of 0.8 for every 10 dB of hearing loss. Comparisons with data from previous studies of upward spread of masking, and assumptions about underlying physiological mechanisms, led to the conclusion that more gradual than normal growth-of-masking slopes reflect larger (steeper) growth-of-response slopes at the probe frequency in regions of hearing loss. Derived response-growth exponents increased by a factor of 1.2 for every 10 dB of hearing loss (HL), from an exponent around 0.25 at 0 dB HL to an exponent around 1.0 at 75 dB HL (linear response growth). Masking sensitivity constants (kappa), the minimum effective masker levels, indicated that masking began at slightly higher masker levels in subjects with sensorineural hearing loss than in subjects with normal hearing. It was concluded that higher masked thresholds in regions of hearing loss were due primarily to a loss of active gain at the probe frequency and were not due to an excessive response at the probe frequency to the lower-frequency masker. For ON-frequency masking conditions, growth-of-masking slopes were not different from normal in hearing-impaired subjects. ON-frequency masking began when the effective power within an auditory filter at the probe frequency reached elevated absolute threshold at the probe frequency. Critical ratios were normal except for one subject with the most hearing loss.

Characterizing Changes in Auditory Filter Bandwidth as a Function of Level

Informational masking (IM) refers to elevations in signal threshold caused by masker uncertainty. The purpose of this study was to investigate two factors expected to influence IM in hearing-impaired listeners. Masked thresholds for a 2000-Hz signal in the presence of simultaneous multitone maskers were measured in 16 normal-hearing (NH) and 9 hearing-impaired (HI) listeners. The maskers were 70 dB SPL average total power and were comprised of fixed-frequency components between 522 and 8346 Hz that were separated from each other by at least 1/3 oct and from the signal by at least 2/3 octs. Masker uncertainty was manipulated by randomly presenting each masker component with probability p = 0.1,0.2,...,0.9, or 1.0 across different trial blocks. Energetic masking was estimated as the amount of masking for p = 1.0, where masker uncertainty was minimum. IM was estimated as the amount of masking in excess of energetic masking. Decision weights were estimated by a regression of the listener's yes/no responses against the presence or absence of the signal and masker components. The decision weights and sensation levels (SLs) of the stimulus components were incorporated as factors in a model that predicts individual differences in IM based on the level variance (in dB) at the output of independent auditory filters [Lutfi, J. Acoust. Soc. Am. 94, 748-758 (1993)]. The results showed much individual variability in IM for the NH listeners (over 40 dB), but little IM for most HI listeners. When masker components were presented to a group of NH listeners at SLs similar to the HI listeners, IM was also similar to the HI listeners. IM was also similar for both groups when the level per masker component was 10 dB SL. These results suggest that reduced masker SLs for HI listeners decrease IM by effectively reducing masker variance. Weighting efficiencies, computed by comparing each listener's pattern of weights to that of an ideal analytic listener, were a good predictor of individual differences in IM among the NH listeners. For the HI listeners weighting efficiency and IM were unrelated because of the large variation in masker SLs among individual listeners, the small variance in IM, and perhaps because broadened auditory filters in some listeners increased the covariance in auditory filter outputs.

The development and decline of forward masking

When a short signal tone is presented after a noise or tone masker, the threshold for detecting the signal is raised the smaller the gap duration between the masker and the signal is. This phenomenon is termed forward masking and refers to the fact that a masker affects the signal threshold when both are presented in a non-simultaneous, consecutive manner. With increasing temporal separation, signal threshold usually drops to performance in silence when the gap is in the region of hundreds of milliseconds. As a possible explanation for forward masking, mainly two different mechanisms have been discussed in the literature: (i) continuation or persistence of neural activity (e.g., Plomp 1964; Oxenham and Moore 1994), referring to temporal integration of neural activity at presumably higher stages than the auditory nerve; (ii) neural adaptation (e.g., Duifhuis 1973; Nelson and Swain 1996), assuming adaptation at various levels of the auditory pathway (including high levels). A third possible source for interaction of masker and signal is linked to the ringing of the auditory filters but is generally assumed to be negligible for signal frequencies of 1 kHz or higher (e.g., Vogten 1978). It is still unclear whether temporal integration or adaptation can better account for forward masking in various stimulus configurations (Oxenham 2001), nor have both mechanisms been compared directly in a common modeling framework to investigate their relation. The current study compares two well established models of temporal processing in the auditory system using a unified modeling framework: (i) the temporal-window model (e.g., Oxenham and Moore 1994) representing a temporal-integration mechanism and (ii) the adaptation-loop model (e.g., Dau et al. 1996) as the representative for the adaptation mechanism. The unified modeling framework shares a compressive, non-linear auditory filter stage and a template-based (optimal detector) decision stage. The question is, whether the temporal-window model and the adaptation-loop model can be considered

In a recent Geophysics Letter, Eisner et al. (2013) measure peak frequencies in velocity spectra of direct waves from microseismic events. The authors interpret these peak frequencies as being caused by attenuation and use them for estimating the effective Q -factor of the medium. Nevertheless, I point out a significant problem with these results, namely, that the authors’ model for the earthquake source is not accurate enough to allow the determination of Q for the surrounding medium. This can be seen by considering the following question: Do the Q values reported in Eisner et al. (2013) characterize the medium, the source model, or the spectral data for direct arrivals? Eisner et al. (2013) assume the first of these answers to be correct; however, careful analysis shows that the correct answer is actually the last one. In answering the above question, note that the Q and t * values measured by this method are not unique physical properties of the medium because they change whenever a different source model is used. The flat plateau at f f c in the Eisner et al. (2013) source spectrum is unrealistic, and yet this plateau is the principal tool of the authors’ method for Q estimation. If we use, for example, Brune’s (1970) model as a more accurate one, then the Q would be much larger than the values reported by Eisner et al. (2013) (see the subsection “Meaning of Q ” below). Thus, the inferred Q s follow from the assumed source model. However, these Q values still cannot be viewed as corrections to the source model because even after these corrections, the model fits the observations very poorly (see the subsection “Data fit”). Therefore, the Q values proposed by Eisner et al. (2013) can only be interpreted as the observed spectral-peak frequencies heuristically multiplied by traveltimes: …

The speech intelligibility index (SII) model was modified to allow individualized parameters. These parameters included the relative weights of speech cues in five octave-frequency bands ranging from 0.25 to 4 kHz, i.e., the band importance function, and the transfer function that allows the SII to generate predictions on speech-recognition scores. A Bayesian adaptive procedure, the quick-band-importance-function (qBIF) procedure, was utilized to enable efficient estimation of the SII parameters from individual listeners. In two experiments, the SII parameters were estimated for 30 normal-hearing adults using Institute of Electrical and Electronics Engineers (IEEE) sentences at speech levels of 55, 65, and 75 dB sound pressure level (in Experiment I) and for 15 hearing-impaired (HI) adult listeners using amplified IEEE or AzBio sentences (in Experiment II). In both experiments, even without prior training, the estimated model parameters showed satisfactory reliability between two runs of the qBIF procedure at least one week apart. For the HI listeners, inter-listener variability in most estimated SII parameters was larger than intra-listener variability of the qBIF procedure.

Speech material influences the relative contributions of different frequency regions to intelligibility for English. In the current study, whether a similar effect of speech material is present for Mandarin Chinese was investigated. Speech recognition was measured using three speech materials in Mandarin, including disyllabic words, nonsense sentences, and meaningful sentences. These materials differed from one another in terms of the amount of contextual information and word frequency. The band importance function (BIF), as defined under the Speech Intelligibility Index (SII) framework, was used to quantify the contributions across frequency regions. The BIFs for the three speech materials were estimated from 16 adults who were native speakers of Mandarin. A Bayesian adaptive procedure was used to efficiently estimate the octave-frequency BIFs for the three materials for each listener. As the amount of contextual information increased, low-frequency bands (e.g., 250 and 500 Hz) became more important for speech recognition, consistent with English. The BIF was flatter for Mandarin than for comparable English speech materials. Introducing the language- and material-specific BIFs to the SII model led to improved predictions of Mandarin speech-recognition performance. Results suggested the necessity of developing material-specific BIFs for Mandarin.

Behavioral measures of spatial selectivity in cochlear implants are important both for guiding the programing of individual users’ implants and for the evaluation of different stimulation methods. However, the methods used are subject to a number of confounding factors that can contaminate estimates of spatial selectivity. These factors include off-site listening, charge interactions between masker and probe pulses in interleaved masking paradigms, and confusion effects in forward masking. We review the effects of these confounds and discuss methods for minimizing them. We describe one such method in which the level of a 125-pps masker is adjusted so as to mask a 125-pps probe, and where the masker and probe pulses are temporally interleaved. Five experiments describe the method and evaluate the potential roles of the different potential confounding factors. No evidence was obtained for off-site listening of the type observed in acoustic hearing. The choice of the masking paradigm was shown to alter the measured spatial selectivity. For short gaps between masker and probe pulses, both facilitation and refractory mechanisms had an effect on masking; this finding should inform the choice of stimulation rate in interleaved masking experiments. No evidence for confusion effects in forward masking was revealed. It is concluded that the proposed method avoids many potential confounds but that the choice of method should depend on the research question under investigation.

The threshold of a pulsed sinusoid centered in a notched-noise masker was measured as a function of the bandwidth of the notch. Assuming that the amount of masking reflects the amount of noise passing through an auditory filter centered on the signal frequency, the derivative of a curve fitted to the data is an estimate of the auditory filter shape [R. D. Patterson, “Auditory filter shape,” J. Acoust. Soc. Am. 55, 802–809 (1974)]. Estimated of the bandwidth of the auditory filter, defined as the half-power bandwidth, were found for three frequencies (1000, 2000, and 4000 Hz) and five noise spectrum levels (10, 20, 30, 40, and 50 dB). The bandwidth of the auditory filter increases with masker intensity above 30-dB noise spectrum level. At these higher intensities, an interaction with signal frequency is also observed, the bandwidth being relatively greater at higher frequencies. [This research was supported by the NIH.]

In psychophysical experiments, the reduction in the effect of one stimulus (the masker) due to the addition of a second stimulus (the suppressor) is attributed to the suppression of the masker by the suppressor. Kramer [reported by F. L. Wightman, T. McGee, and M. Kramer, in Psychophysics and Physiology of Hearing, edited by E. F. Evans and J. P. Wilson Academic, London, 1977)] measured “unmasking patterns” (signal threshold as a function of suppressor frequency) in listeners with high-frequency hearing loss. Although he observed normal suppression effects in a region of normal hearing, he found no evidence of suppression in regions of impaired heating. One factor which might have contributed to this failure was an unlucky selection of the suppressor level. This experiment compares the effect of suppressor level in normal and hearing-impaired frequency regions. Thresholds for 10-ms signals (10-ms, cosine-squared ramps with no steady state) were obtained for signals at 1 and 4 kHz forward masked by 300-ms, narrow-band (50-Hz-wide) noise maskers centered at the signal frequencies. Masker levels were 50 and 70 dB (total noise power), respectively. Suppressor levels varied from 30 to 90 dB SPL; suppressor frequencies were 0.7 kHz for the 1-kHz masker and 2.8 and 5.0 kHz for the 4-kHz masker. As observed by Kramer, suppression effects appear to be absent in regions of hearing impairment. However, in this experiment, this absence does not appear to be correlated with an absence of a difference between simultaneous and forward masking. [Research supported by grants from NSF and NIH.]