Method and System for Assessing Supra- Threshold Hearing Loss

STATEMENT OF GOVERNMENT RIGHTS

This invention was made with Government support under Contract No. N00014-09-1-

0073 awarded by the Department of the Navy, and Contract No. DC013027 awarded by the National Institutes of Health. The Government has certain rights in the invention.

BACKGROUND

The present invention relates to methods and systems for assessing and compensating for hearing loss.

Recent basic research in animals as well as in human listeners suggests that noise exposure and associated accelerated auditory aging leads to cochlear neuropathy, which can produce supra-threshold hearing deficits without any elevation of hearing thresholds. This supra-threshold hearing loss (STHL), currently undiagnosed with standard audiometric tests, causes degraded encoding of amplitude modulation and leads to poor speech understanding, especially in noisy environments.

As many as 5-15% of people seeking audiological care are found to have normal hearing thresholds (i.e., can detect the presence of quiet sounds as easily as healthy young adults). Such listeners typically seek help specifically because they have trouble

understanding speech in social settings (e.g., group conversations in cocktail parties). These communication difficulties, caused by the undetected supra-threshold hearing loss, can lead to social isolation, withdrawal, and depression. Current approaches to aid such listeners are unsatisfactory and do not take into account the form of the underlying perceptual deficits, in part because their root cause is only now coming to light. In addition, a common complaint of hearing aid users is that their hearing aids do not allow them to converse in restaurants or other noisy settings. Many of these listeners may suffer from a mixture of traditional hearing loss and STHL, but current hearing aid processing schemes focus only on amplifying sound to overcome elevated thresholds. For a listener with STHL, such amplification can decrease the perceptual salience of information-conveying amplitude modulation; thus current amplification strategies are likely to interfere with rather than enhance speech understanding in listeners with STHL. The current state of the art in the diagnosis of hearing loss focuses on dysfunction of the cochlea; treatments focus on amplifying sound to overcome elevated hearing thresholds, which result when fragile cochlear structures are damaged and fail to amplify faint sounds. However, recent work reveals that there is widespread "hidden" (i.e., undetected by conventional testing) hearing loss caused by a loss of auditory nerve fibers carrying sound from the cochlea to the brain, the result of both noise exposure and aging.

SUMMARY

The disclosure is directed to novel individualized hearing-aid signal processing strategies to overcome "hidden" (i.e., undetected using conventional testing) hearing deficits. The research community is beginning to recognize the prevalence and importance of such hidden hearing loss, which arises from a loss of ascending auditory nerve fibers (cochlear neuropathy) and results in degraded speech understanding in noisy settings. However, this knowledge has not yet reached the broad clinical community, and there are no effective clinical interventions or treatments to ameliorate the effects of neuropathy.

A diagnostic approach is used to quantify the extent of cochlear neuropathy in individual listeners, as well as computational models to predict the perceptual consequences of different patterns of nerve-fiber loss. In addition, behavioral and electrophysiological measures are used to identify listeners with STHL and to characterize their loss. From the quantitative predictions, an algorithm may be used to improve, on an individual basis, perception of speech in noisy settings, and parameters can be established for individualized hearing-aid processing to improve a listener's ability to understand speech in noise. The diagnostic measures and individualized processing strategy can benefit listeners with STHL and mixtures of traditional hearing loss and STHL.

In one aspect, a diagnostic method is disclosed of producing user-specific, per-band adjustment values for configuring into a hearing aid for use by a hearing aid user. The adjustment values reflect temporal coding fidelity as represented by an envelope-following response (EFR) of the hearing aid user, and are used by processing circuitry of the hearing aid to adjust per-band envelope amplitudes of an input acoustic speech signal to generate a modified acoustic speech signal provided to the hearing aid user. The method includes:

applying an auditory stimulus to the user over a test period having a plurality of epochs, the auditory stimulus including a plurality of carrier signals in respective auditory frequency bands corresponding to respective distinct cochlear sections, the carrier signals being applied at super-threshold levels above a hearing threshold of the hearing aid user and being amplitude-modulated by respective modulation signals of respective distinct modulation frequencies less than 500 Hz;

capturing electromagnetic signals generated by auditory system activity of the hearing aid user throughout the test period; and

converting the electromagnetic signals into an electronic representation and processing the electronic representation. The processing includes:

(i) computing a spectrum for each of a plurality of epochs of the test period, and computing an average spectrum across all epochs;

(ii) sampling the average spectrum to identify peaks at spectral frequencies corresponding to the distinct modulation frequencies of the amplitude-modulated carrier signals; and

(iii) computing per-band EFR strengths as either amplitudes or phase-locking values for the peaks at the spectral frequencies, the amplitudes or phase-locking values being mapped to respective ones of the per-band adjustment values for the hearing aid.

In another aspect, a hearing aid is disclosed that includes an input sound transducer to receive an input acoustic speech signal, an output sound transducer to provide a modified acoustic speech signal to a hearing aid user, and electronic circuitry coupled to the input sound transducer and the output sound transducer for processing the input acoustic speech signal to produce the modified acoustic speech signal. The electronic circuitry includes signal-processing circuitry configured and operative to:

generate an electronic representation of the input acoustic speech signal and divide the electronic representation into a plurality of auditory frequency bands to produce a corresponding plurality of per-band signals, the auditory frequency bands corresponding to respective distinct cochlear sections;

extract, from the per-band signals, respective temporal envelopes having respective modulation frequencies less than 500 Hz and respective envelope amplitudes;

apply user-specific, respective per-band adjustments to the envelope amplitudes of the per-band signals to obtain modified per-band signals having respective temporal envelopes of respective adjusted envelope amplitudes, the user- specific per-band adjustments using respective user-specific adjustment values configured into the hearing aid specifically for the hearing aid user and reflecting measurements of temporal coding fidelity for the hearing aid user; and

synthesize and apply to the output sound transducer an electronic representation of the modified acoustic speech signal by combining the modified per- band signals.

BRIEF DESCRIPTION OF THE DRAWINGS

The foregoing and other objects, features and advantages will be apparent from the following description of particular embodiments of the invention, as illustrated in the accompanying drawings in which like reference characters refer to the same parts throughout the different views.

Figure 1 is a block diagram of a diagnostic system for assessing supra-threshold hearing loss of a listener;

Figure 2 is a block diagram of signal processing applied to EEG signals according to a disclosed diagnostic method;

Figure 3 is a set of plots of envelope-following response (EFR) versus noise configurations in different frequency bands;

Figure 4 is a set of plots including (a) speech-in-noise threshold versus high-band EFR, and (b) phase-locking value (PLV) versus frequency;

Figure 5 is a schematic depiction of a derived-band analysis method.

DETAILED DESCRIPTION

The current clinical standard of care for screening peripheral hearing is to determine the faintest sound that a subject or listener can detect at each of a range of acoustic frequencies. Such "threshold" measures do nothing to measure the fidelity with which clearly audible sound (such as conversational speech) is encoded within the subject's auditory processing organs. While more in-depth testing is sometimes undertaken, these tests (e.g., word recognition, speech in noise, etc.) do little to diagnose the root cause of any deficits that are found; thus, they do not identify any method to compensate for these problems.

When it comes to treatment, current hearing aid signal-processing strategies are designed to amplify sounds to improve audibility, thereby compensating for elevated hearing thresholds. Noise suppression algorithms, directional microphones and audio inputs from auxiliary devices are incorporated in many devices in an attempt to improve signal-to- noise ratio at the ear of the listener; however, those that work best require additional paraphernalia that most users are loathe to use, and none are individualized to the extent proposed here. Indeed, the amplification that is the basis of most current hearing-aid technologies can decrease the perceptual salience of information-conveying amplitude modulation for a listener with STHL; current amplification strategies are likely to interfere with rather than enhance speech understanding in listeners with STHL.

The need to develop diagnostic measures and treatment plans for STHL goes beyond the population with normal threshold. Listeners with clinical hearing loss can also suffer from STHL. About 20 percent of adults in the United States (48 million) report some degree of clinical hearing loss (defined by elevated thresholds). However, nearly 23 million adults in the United States have a clinical hearing loss but do not use hearing aids. Many people fit with hearing aids opt not to wear them. One of the top reasons is a lack of benefit, especially in noisy environments (e.g., the hearing aid user at the restaurant complains that "all the aid does is amplify the sound of the dishes clanking in the background"). The lack of benefit in current hearing aids likely comes from the lack of compensation for the specific deficits caused by STHL.

One key aspect of the disclosed methods and apparatus is a novel diagnostic approach that can identify and quantify specific deficits caused by STHL, leading to corresponding personalization of a compensating hearing aid algorithm that compensates for the specific deficits caused by a listener's specific pattern of hidden hearing loss.

Specifically described are diagnostic measures of the fidelity with which supra- threshold sound is encoded at different carrier frequencies, which enable mapping of a listener's "temporal coding profile" (TCP). Using the TCP, it is possible to predict a listener's ability to perceive acoustic amplitude modulations that convey pitch features, interaural timing cues in high frequencies, and speech meaning (depending on the frequencies of the amplitude modulations). Each of these features is critical for communicating effectively in everyday settings, particularly when there are competing sounds in the environment— the very settings where most current amplification-based hearing aids fail to provide significant benefit to listeners. The disclosed diagnostic approach can enable rapid identification of listeners who suffer from STHL and who are likely to benefit from the novel hearing aid processing strategies. A proposed signal-processing algorithm focuses on restoring the specific degradations in auditory coding of modulation caused by STHL in order to improve auditory perception. The signal-processing algorithm takes advantage of the new measurement approaches that map out each individual's TCP as a function of acoustic carrier frequency (place of excitation along the cochlea). Using these individualized assessments, individualized signal-processing schemes are created that compensate for subject-specific coding deficits by maximizing information transfer of key acoustic modulations important in sound segregation and speech perception. The combination of novel diagnostic measures and individualized optimization of modulation transfer can improve speech understanding for listeners with STHL as well as those with a mixture of traditional hearing loss and STHL. The hearing aid algorithm can aid listeners who have trouble communicating in crowded settings and who gain little benefit from conventional hearing aids, especially listeners with normal thresholds or with mild to moderate hearing loss.

Specific advantages include:

1. The diagnostic measures can identify and quantify currently undiagnosed hidden hearing loss, including determining the fidelity of auditory coding as a function of frequency. While there are existing clinical tests to identify listeners who have trouble understanding speech in noise despite having normal thresholds, none address the underlying mechanisms causing the loss and none have the specificity of the technique described herein.

2. The signal processing algorithm can improve speech understanding in noisy settings for listeners with hidden hearing loss. In contrast, current hearing aid algorithms, which generally depend upon amplifying sound, may interfere with speech understanding for listeners with hidden hearing loss and normal hearing thresholds. For these listeners, increasing sound levels through amplification may further degrade how well amplitude modulation in already audible sound is represented in the brainstem.

3. The hearing aid processing is compatible with current hearing aids and does not require additional hardware.

The proposed technique contains two parts: 1) Diagnostic measurements to identify and quantify the deficits caused by supra-threshold hearing loss (STHL) by measuring a subject's temporal coding profile at a range of acoustic carrier frequencies, 2) A personalized signal processing algorithm to compensate for a subject's pattern of temporal coding deficits at different acoustic frequencies, with parameters determined by the subject's temporal coding profile. The subject-specific parameters may be used to customize operation of a hearing aid to tailor its performance to the subject's temporal coding profile, improving the subject's experience. The general approach to each of these is described below. Diagnostic measurement of temporal coding

An electrophysiological measurement scheme is used to identify a subject's auditory brainstem response by detection and processing of electromagnetic signals obtained directly from the subject, such as through electroencephalography (EEG). A representative setup for EEG-based measurement of auditory brainstem response is shown, for example, in US patent publication 2006/0120538 (e.g., Figure 4).

In particular, envelope following responses (EFRs) are measured in response to tones presented at an intensity that is well above threshold. Testing includes obtaining samples of an envelope modulation amplitude over several epochs of a testing session, and the results are analyzed to derive a measurement of EFR strength of the listener. For example, there may be 2000 300-millisecond epochs over a 25-minutes long test session, where an epoch is the time duration of one trial with a particular auditory stimulus and corresponding EEG response. An EFR device presents auditory stimuli designed to assess temporal coding fidelity to the listener as a function of frequency. Discrete sets of acoustic signals, having different fundamental frequencies (FOs) and that stimulate different cochlear frequency regions, are presented simultaneously to obtain multiple EFRs simultaneously. Background noises may be included separately in each frequency band to assess the robustness of temporal coding against noise. A characteristic of the EFR amplitude at each frequency is used to quantify suprathreshold coding fidelity in the corresponding frequency band. Listener's EEG responses are recorded by the EFR device during the sound presentation. Processing circuitry analyzes the EEG data and generates metrics to quantify suprathreshold coding fidelity for each frequency region.

Figure 1 is a block diagram of a system for the EEG measurements. It includes a source 10 of auditory stimuli that are provided to a listener 12, who is also referred to as a "subject" or "user" herein (e.g., hearing aid user). An electrical response of the listener 12 in the form of EEG signals is acquired by EEG signal acquisition block 14, which provides EEG signal data to an EEG signal analysis block 16. The EEG signal analysis block 16 generates output signals or data representing the specific coding profile of the listener 12, and/or listener-specific parameters that may be used in a compensation device such as a hearing aid, as described further below. There are two EEG electrodes placed on respective earlobes of the listener 12, one on each side. The EEG signals from these two channels are averaged and the averaged signal is subtracted out from the EEG signal recorded from each of the other EEG channels.

According to one embodiment of a method of obtaining diagnostic measurements, a set of relatively narrowband tones is used for the auditory stimuli. In one example, the auditory spectrum is divided into three frequency bands, each containing a set of harmonic tone components with a respective fundamental frequency (F0), thus resulting in per-band amplitude-modulation at a respective distinct modulation frequency as follows:

# Band FO/Modulation frequency

1. <1500 Hz 114 Hz

2. 1500-3500 Hz 170 Hz

3. >3500 Hz 236 Hz

Generally, the modulation frequencies are below 500 Hz, which represents a practical cutoff of transmission of auditory-related electromagnetic signals within the cranium. Those skilled in the art will appreciate that different numbers of bands and different specific modulation frequencies may be used.

In the diagnostic measurement, the complex tones in three frequency bands are presented simultaneously. Each trial is 300ms long. The inter-trial interval is preferably random and uniformly distributed between 300 and 400 ms, jittered from trial to trial to ensure that EEG noise unrelated to the stimulus occurs at a random phase between -π and π for frequencies above 10 Hz. The multi-band complex tones are presented at super-threshold level (>76 dB SPL) in one of two noise configurations: (1) no noise, (2) noise in all bands. As shown below, noise in other bands does not adversely affect the frequency specificity of the measurement in a given band. Each type of trial is presented a large number of times, e.g., on the order of 1000 times. For a randomly selected set of the trials, e.g., 500 out of 1000 trials, the polarity of the stimulus was inverted. This enables separation of the responses phase- locked to the modulation envelope from the responses phase-locked to the temporal fine- structure of the acoustic inputs. Different type of trials may be intermingled and presented in random order. For each subject, the recording session of the diagnostic measurement may last about 25 minutes for presenting a total of 2000 trials. Throughout data collection, participants may engage in quiet activity such as watching a silent, captioned movie of their choice, ignoring the acoustic stimuli.

Figure 2 shows details of the EEG signal analysis 16. EEG signal data are provided to highpass filter (HP filter) 20 for 70-Hz highpass filtering to retain subcortical responses while minimizing line noise, low-frequency noise and interference from cortical activity. The highpass filtered signals are applied to epoch block 22 where they are broken into epochs, from 0 to 300 ms relative to the onset of each 300-ms-long complex tone. Outlier epochs may be rejected, such as those having peak-to-peak deflections exceeding some maximum, to reduce artifacts from spurious activity such as eye blinks, etc. For remaining epochs, the raw EEG response is referenced to an average of the two earlobe channels. In block 24, the spectrum of the EEG signals is computed, such as by using a fast Fourier transform (FFT), and per-band spectrum amplitude values are provided to respective blocks 26 (26-1, ... , 26- N) for computation of respective phase-locking values (PLVs), PLV-1, ... , PLV-N. These are provided to respective blocks 28 (28-1, ... , 28-N) for computation of respective Fb values Fb i, ... , Fb _N , which are described below.

Epochs may be separately analyzed corresponding to different noise configurations. Different metrics can be used to characterize the strength of the EFR, including the raw amplitude and the phase-locking value (PLV). While both metrics can be used directly to characterize EFR strength, they may also be calculated as differential values representing a difference between one auditory response for a signal-only acoustic input and a second auditory response for an acoustic input having signal and noise. The EFR strength metric for the two responses can be subtracted to yield a derived metric that can serve as the derived value of the analysis. This approach ensures that some nuisance variables in the measurement are reduced, producing values that are more directly comparable across subjects.

In one embodiment, the strength of the EFR is characterized by the spectral amplitude for each band, computed separately for each subject, for each group of epochs. Measurements over many epochs in a test period are per-band averaged/combined to compute per-band amplitude values.

The multi-band design and the super-threshold presentation level substantially reduce neural responses from off-frequency regions on the cochlea, thus providing a frequency- specific approach to assess super-threshold coding fidelity. Key aspects of this approach include (1) its design to measure super-threshold hearing loss, which is different from clinical hearing loss (threshold elevation) and associated diagnostic methods. The sound level in the disclosed approach is much higher than those used in prior studies. Aspects also include (2) efficiency by stimulating multiple frequency bands simultaneously, and (3) use of complex tones in one frequency band that have a strong masking effect on responses to stimuli in other frequency bands, thus providing frequency-specific EFR results more reliably.

Figure 3 illustrates certain group average data that illustrates the frequency specificity of the diagnostic measurement technique. In Figure 3 the three bands are identified as Low (L), Mid (M) and High (H), and for each band there is a plot of EFR amplitude for several noise scenarios:

L - noise in Low band only

M - noise in Mid band only

H - noise in High band only

LM - noise in both Low and Mid bands

LH - noise in both Low and High bands

MH - noise in both Mid and High bands

LMH - noise in all three bands

Figure 3 shows that EFR in one frequency band is only affected by adding noise in that same frequency band, and does not change when adding noise in other frequency bands. Because of this feature, in practice it is only necessary to use one trial type with noise in all bands, in addition to the no-noise trial type. Although not shown in Figure 3, the frequency specificity of the measurement is also indicated by other results showing that EFR strengths in Low band and in High band are not correlated across subjects.

Figure 4 presents additional results. The plot in upper part A shows that the High- band EFR strength may predict subjects' performance in a speech-in-noise test. Based on this correlation, a threshold in EFR strength can be chosen to identify subjects with super- threshold hearing loss. In the illustrated example, all subjects with EFR strength weaker than 1 dB perform poorer than the average level in the "good" listener group. Lower part B presents EFR responses (as PLVs) from three example subjects, showing that EFR pattern across frequency bands can be very different for different subjects. This further underscores the necessity to use a frequency specific measurement to assess super-threshold hearing loss.

Besides providing a diagnosis for the super-threshold hearing loss, the disclosed measurement technique can also provide a quantitative measure of the super-threshold coding fidelity in a frequency specific manner, which can be incorporated into any of a variety of STHL-compensating signal processing algorithms to enhance speech perception. Measured per-band EFR strength can be used in these algorithms to adjust the amount of enhancement applied to the speech signal. Specifically, lower EFR strength indicates a higher degree of deficits in coding sound envelopes, which would generally indicate a need for greater enhancement. For example, some algorithms extract the sound envelope and raise the envelope to some exponent greater than 1 to enhance the envelope. For these algorithms, it is possible to scale the PLVs from the disclosed measurement technique by a suitable scaling or mapping function to yield the exponent parameter values in a desired normalized range. The target normalized range may have a maximum between 1 and 2, so the scaling is selected to map the PLVs accordingly. In this example, a low PLV would be mapped to a parameter value close to 2 (a high degree of enhancement), while a high PLV would be mapped to a parameter value close to 1 (little enhancement). The scaling or mapping function should be monotonic over the range of values, but need not be linear.

In providing a detailed assessment of super-threshold hearing loss in multiple frequency bands, the disclosed diagnostic measurement technique generally requires significantly longer time than typical audiological tests. To reduce screening time in clinical settings, it may be desirable to first run a quick, non-frequency-specific test to screen for general super-threshold hearing loss. Some currently available audiological tests, such as the auditory brainstem response (ABR) and the middle ear muscle reflex (MEMR), have the potential to serve as a quick assay. Once the screening suggests that a super-threshold hearing may be present, the disclosed diagnostic measurement can be used to obtain a detailed picture of the hearing loss along the cochlea.

Personalized signal processing algorithm

One of the main deficits caused by STUL is a degradation in the neural representation of (and subsequent loss of perceptual sensitivity to) envelope modulations. A signal processing algorithm is used to enhance the information-conveying modulations in the acoustic signal in order to compensate for the degradation caused by STUL. The algorithm comprises two components corresponding to enhancement in two envelope-frequency ranges. The algorithm may be implemented in electronic circuitry of a hearing aid or similar device. Generally, a hearing aid includes an input sound transducer (e.g., microphone) to receive an input acoustic speech signal, an output sound transducer (e.g., loudspeaker) to provide a modified acoustic speech signal to a hearing aid user, electronic circuitry coupled to the input sound transducer and the output sound transducer for processing the input acoustic speech signal to produce the modified acoustic speech signal, the electronic circuitry including signal -processing circuitry configured and operative according to the signal processing algorithm. The signal-processing circuitry may include a programmed digital signal processor executing a set of program instructions that implement the algorithm.

A. Slow modulation enhancement (<32 Hz)

An envelope enhancement procedure aimed at slow modulations may be utilized. The original acoustic speech signal is first divided into narrow bands using band pass filters (e.g., 3 bands using 3rd order Butterworth with cutoff frequencies of 50-1500; 1500-3000; 3000- 8000 Hz). The temporal envelopes E(t) are extracted from each band by first full-wave rectifying and then low-pass filtering with a cutoff frequency of 32 Hz. The extracted envelope in a given band is manipulated to maximize the subject's perception of the slow modulations in that band, based on personalized fittings (e.g., derived from the measurement procedure described above). Specifically, the instantaneous envelope amplitude value of the envelope of band b, denoted as (Eb(t)), is monotonically remapped by raising the value to the power Kb(t), where Kb(t) ranges from a maximum expansive value Kmax (e.g., Kmax=4) to a compressive value Kmin (e.g., Kmin =0.3). The expression for Kb(t) is computed as:

(E _v (t)-E _min ) where Emin is the minimum envelope value. For each subject, the temporal coding fidelity measured for each frequency region is used to compute a listener-specific

compensating scaling factor Fb which further sharpens the envelope modulation to compensate for the coding deficit caused by listener's STHL. Fb is actually a set of per-band values Fbi, ... , FbN as shown in Figure 1.

B. Pitch enhancement

After the slow modulation in the signal is enhanced, a pitch enhancement algorithm may be used to further compensate for the degradation in the temporal modulation at the fundamental frequency of the speech, or the pitch frequency. A short-time frequency decomposition is performed to transform the signal into time-frequency bins. Within each time-frequency bin, the periodicity of the signal falling in the pitch range is analyzed. If a dominant pitch can be identified in a given time-frequency bin, the algorithm enhances the pitch information using a comb filter. If not, the spectro-temporal content is left unprocessed. A comb filter technique may be used to suppress the spectral energy between harmonic frequencies of the pitch while keeping the spectral envelope intact. The signal is then resynthesized by inverse transforming the resulting time-frequency content and summing the results.

Diagnostic Measurement - Alternative Embodiment

According to a second embodiment, envelope following responses (EFRs) are measured in response to broadband complex tones consisting of harmonics of a fundamental frequency (F0) in the pitch range (e.g., pure tones that are harmonics of 100 Hz), presented at an intensity that is well above threshold. A derived-band subtraction technique is used to obtain EFRs originated from different frequency regions of the cochlea.

Derived-band EFRs are obtained by subtracting EFR recordings acquired under different highpass noise masking conditions, as described more below. Using this method, derived-band responses are obtained for adjacent frequency regions (e.g., 0-1.5 kHz, 1.5-3 kHz, 3-6 kHz, 6-10 kHz).

Figure 5 illustrates this technique for the 3-6 kHz band. The two responses are labelled A and B, with response A having a 6-kHz masking cutoff and response B having a 3- kHz masking cutoff as shown. Response B is subtracted from response A, and the difference is the 3-6 kHz derived-band response.

The spectral level of the highpass noise is chosen so that the corresponding broadband noise with same spectral level completely eliminates the EFR generated by the complex tone (e.g., 70 dB SPL). The EFRs are measured in highpass conditions (e.g., cutoff frequencies: 1.5 kHz, 3 kHz and 6 kHz). In each highpass condition, a lowpass noise with the same cutoff frequency as the highpass noise is added to the stimulus. While the spectral level of the highpass noise is fixed throughout the measurement, the spectral level of the lowpass noise for each token is randomly selected from a range of levels (e.g., [65, 60, 55, 50 dB]) with equal probability. For each combination of highpass cutoff frequency and lowpass noise level, multiple long tokens are presented (e.g., twenty 10-second tokens) in random order. The interstimulus interval varies uniformly (e.g., between 1 and 1.2 seconds) to remove artifacts associated with a fixed repetition rate.

After the narrowband EFR is computed by subtracting two adjacent highpass responses (Figure 5), the EFR power is estimated in the frequency domain using a complex principal component analysis approach to combine measurements across channels. This approach optimally adjusts for phase disparities in the signal across channels, which improves the S R of the extracted EFR significantly. A multichannel estimate of the phase- locking value (PLV) is used to determine whether or not each EFR peak exceeds the measurement noise floor. The PLV is convenient for this purpose because the noise floor distribution is independent of background noise levels. The slope of the EFR amplitude at FO with lowpass noise level is used to quantify suprathreshold coding fidelity as represented by the strength of the listener's EFR , computed by fitting a straight line to the measurements as a function of the lowpass noise levels.

While various embodiments of the invention have been particularly shown and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.