Assessing the benefit of acoustic beamforming for listeners with aphasia using modified psychoacoustic methods.

Acoustic beamforming has been shown to improve identification of target speech in noisy listening environments for individuals with sensorineural hearing loss. This study examined whether beamforming would provide a similar benefit for individuals with aphasia (acquired neurological language impairment). The benefit of beamforming was examined for persons with aphasia (PWA) and age- and hearing-matched controls in both a speech masking condition and a speech-shaped, speech-modulated noise masking condition. Performance was measured when natural spatial cues were provided, as well as when the target speech level was enhanced via a single-channel beamformer. Because typical psychoacoustic methods may present substantial experimental confounds for PWA, clinically guided modifications of experimental procedures were determined individually for each PWA participant. Results indicated that the beamformer provided a significant overall benefit to listeners. On an individual level, both PWA and controls who exhibited poorer performance on the speech masking condition with spatial cues benefited from the beamformer, while those who achieved better performance with spatial cues did not. All participants benefited from the beamformer in the noise masking condition. The findings suggest that a spatially tuned hearing aid may be beneficial for older listeners with relatively mild hearing loss who have difficulty taking advantage of spatial cues.

INTRODUCTION

It is common for individuals with aphasia—i.e., language impairment resulting from stroke or other neurological injury/disease—to report difficulty understanding speech in noisy environments. The challenge of listening to target speech while ignoring or filtering out background noise, known as the “cocktail party problem” [Cherry (1953); see Middlebrooks for a series of recent reviews], has high relevance for everyday communication, as real-world conversations often take place in settings that are acoustically complex. While the majority of past research on receptive speech processing in persons with aphasia (PWA) has focused on auditory language comprehension in quiet settings, several recent studies have directly investigated the ability of persons with aphasia (PWA) to selectively attend to and understand speech in the presence of auditory maskers [e.g., Rankin and Villard and Kidd (2019)]. These studies have provided evidence that PWA—even, in some cases, PWA with milder aphasia types thought to be characterized primarily by expressive language deficits—require higher target-to-masker ratios (TMRs) than do age-matched controls in order to successfully understand target speech. These laboratory-based findings validate the anecdotal reports from PWA that background sounds can present major challenges in their ability to understand conversational partners in everyday situations.

While it is not yet clear precisely why PWA encounter difficulty in multi-talker environments, there is reason to believe that this difficulty could be due to impairments in cognitive abilities such as selective attention. Aphasia has historically been defined as a disorder that solely affects language abilities, leaving cognitive abilities intact. However, the past several decades have seen an increasing focus in the aphasia literature on impairments in cognitive areas including attention (Hula and McNeil, 2008; Murray, 2000, 2012; Villard and Kiran, 2017), which may be relevant in cocktail party situations where listeners must selectively attend to a target speech stream while ignoring maskers. The observation that PWA may encounter difficulty understanding speech when background sounds are present also gives rise to the question of what types of strategies or approaches might be effective in helping PWA to improve communication in complex sound environments. The identification or development of such strategies, whether rehabilitative or compensatory in nature, is a goal of both clinical significance and practical importance. The primary goal of the current study is to investigate whether a compensatory technique involving acoustic beamforming could improve PWA performance on a masked speech recognition task.

A secondary but related goal of the current study is to explore a possible approach to modifying psychoacoustic methods for use with PWA. Typically, determination of whether a particular rehabilitative approach or compensatory technique is effective for a group of listeners would involve the administration of certain standard speech recognition or psychoacoustic measures. However, it is generally acknowledged that a variety of factors associated with aphasia may make it challenging to obtain valid and reliable measurements from PWA using these tasks, which were developed to test listeners without known cognitive-linguistic impairments. These factors include not only the language deficits that characterize aphasia, but also associated deficits in areas such as reading and verbal working memory. In order to ensure accurate assessment of the benefit of potential strategies for improving speech intelligibility in PWA participants, careful modifications of typical psychoacoustic methods may need to be devised and implemented. Therefore, while the current study's primary focus is on the question of whether a compensatory auditory prosthesis utilizing acoustic beamforming could provide a benefit to some PWA in cocktail party listening situations, it also examines an approach for adapting typical psychoacoustic methods for the purpose of obtaining accurate measurements of speech recognition abilities in PWA.

Energetic and informational masking in aphasia

In measuring the effects of masking and considering possible strategies for improving masked speech recognition, it is important to distinguish between the effects of two broadly defined types of masking: energetic and informational. Energetic masking (EM) refers to reduced neural representation of target sounds due to time-frequency overlap with masker sounds. That is, if the masker energy exceeds the target energy in a given time-frequency region, then the masker likely will dominate the peripheral neural representation of the combined stimulus in that region, making the target difficult or impossible to detect or to identify [e.g., Conroy ]. In that case, performance is adversely affected because the neural representation of the target is “data limited”; i.e., sufficient information from the target source does not propagate up the auditory pathways. However, adequate peripheral neural representation of the target does not ensure the absence of masking. Certain types of listening situations have been found to produce substantial levels of informational masking (IM), or additional masking beyond any EM present in the signal [e.g., Brungart , Freyman , Calandruccio , Holmes ; reviews in Kidd and Mattys ]. IM is thought to result from limitations in the listener's later-stage central processing capabilities and tends to be high in speech-on-speech masking situations where the masker and target are difficult to separate and are easily confused despite adequate audibility of the target [see Kidd and Colburn (2017) for a review]. Susceptibility to IM has been shown to vary considerably from listener to listener, even among young adult listeners with normal hearing, possibly due to inter-individual variability in higher-level cognitive skills such as attention or working memory [e.g., Clayton and Oberfeld and Kloeckner-Nowotny (2016)].

When considering how masking affects speech recognition in PWA, IM may be of particular interest because aphasia produces cognitive-linguistic impairments but does not have any known effect on peripheral auditory function. Our recent study on the respective effects of EM and IM in PWA found that when simple target sentences were masked by speech-shaped, speech envelope-modulated noise—a high-EM, low-IM condition—PWA and age-matched controls performed similarly. However, when the same sentences were masked by intelligible speech—a high-IM, low-EM condition—PWA required significantly higher TMRs in order to comprehend the target sentences (Villard and Kidd, 2019). These results suggest that PWA experience a particular difficulty in perceptually segregating target speech from masker speech and/or selectively attending to target speech. Because the hearing profiles were similar between the two groups in that study, and because only very simple experimental stimuli were presented, the observed group difference likely was not attributable to peripheral factors or to underlying language comprehension deficits. As susceptibility to IM in speech-on-speech masking conditions is thought to be related to higher-level cognitive processes [e.g., Swaminathan and Clayton ], it is hardly surprising that high-IM conditions pose problems for PWA, who by definition have damage to central processing areas.

The finding that PWA may be highly susceptible to the effects of IM could have a variety of implications for these listeners. To begin with, it suggests that PWA may struggle to understand speech in everyday situations where multiple conversational streams are audible; examples might include the family dinner table, a holiday party, the intermission during a play or concert, or even the checkout line at a supermarket. Difficulty understanding a conversational partner in settings like these could adversely affect social relationships, community participation, and quality of life in PWA. Additionally, it is plausible that difficulty filtering out background talkers or other distracting sound sources could mitigate the benefits of language therapy in PWA. Particularly in the earlier stages of recovery, PWA often undergo language therapy in medical settings, many of which may contain substantial background sounds, including the voices of medical personnel, other patients, and visitors, as well as voices from sources such as intercoms and televisions. These auditory environments could therefore contain a substantial amount of IM (Pope, 2010; Pope ). Because background sounds in real-world settings often are difficult to control or to modify, the development of techniques to accurately measure the effects of auditory masking in such environments on PWA, as well as strategies to help PWA better understand target speech, could make a substantial difference in the everyday lives of individuals living with aphasia.

Furthermore, although damage to central processes may be a driving factor behind the challenges faced by PWA in complex listening situations, the possible impact of peripheral factors on the ability to understand speech and the effects of masking is also essential to consider in this population. Aphasia is most common in older individuals, and there has been an increasing recognition of the possibility that many PWA may demonstrate some degree of age-related sensorineural hearing loss (SNHL) in addition to their language deficits (Formby ; Silkes and Winterstein, 2017; Zhang ). The challenge of understanding and addressing the respective contributions of peripheral and central factors to masking susceptibility in individual PWA therefore complicates both the characterization of this problem and the development of possible strategies to address it.

Acoustic beamforming as a possible compensatory strategy for listeners with aphasia

Because the literature on auditory masking in PWA is still quite limited, especially with respect to elucidation of the relative influences of EM and IM, little is currently known about how speech recognition abilities in complex acoustic environments can be improved in this population. A fundamental question in beginning to investigate possible strategies for improving communication is whether a compensatory/prosthetic approach (such as a hearing aid or other amplification system) or a rehabilitative approach (such as auditory training) would be most effective. While commercially available hearing aids designed for listeners with SNHL could provide assistance to PWA in some situations (e.g., improving audibility if hearing loss is present), the potential benefit of standard hearing aids is limited in situations comprising multiple competing talkers in part because they amplify the maskers as well as the target, often providing only modest improvements in TMR. Hearing aids or other amplification systems that include a strong directional component, however, can be more useful in complex auditory environments because they may provide amplification that emphasizes a target source location.

Acoustic beamforming is a highly directional amplification approach that has been found to be effective in laboratory-based studies of speech-on-speech masking in listeners with SNHL [e.g., Kidd ]. The beamforming technology used in past work from our group [e.g., Kidd , Kidd , Best , and Roverud ] consists of an array of spatially distributed, omni-directional microphones worn on the head of a human listener—or, more commonly, the beamforming algorithm is implemented for headphone-based presentation using impulse responses measured while the array is positioned on the KEMAR manikin [see Kidd (2017) for details]. When implemented, the beamformer effectively attenuates sounds originating from locations that are off-axis from a designated target location (usually either directly in front of the listener or at an azimuth specified by eye gaze) and subsequently presents the combined target/masker signal to the listener via a single channel (i.e., monotic or diotic presentation). While the single channel output signal lacks the binaural spatial cues available to listeners in naturalistic, unaided listening situations, its key advantage is an improved TMR that can help boost listener performance on a speech recognition task. Thus, the perceptual segregation of target and masker by relative level is enhanced by the single-channel beamformer; however, this enhancement comes at the cost of the loss of perceptual segregation cues resulting from interaural differences in target and masker waveforms.

In general, the beamforming approach used in the present study has been shown to provide significant benefits for masked speech recognition when there is sufficient spatial separation between target and maskers under certain conditions for both normal hearing (NH) and SNHL listeners (Kidd, 2017; Kidd ), and more recently in cochlear implant users (Yun ). In particular, implementation of the beamformer was found to result in significantly lower (better) speech reception thresholds (SRTs) for both NH and SNHL listeners in a high-EM listening condition where maskers consisted of speech-shaped, speech envelope-modulated noise that was spatially separated from the target (Kidd, 2017). However, in high-IM listening conditions where maskers comprise intelligible speech and are spatially separated from the target, the effect of the beamformer (i.e., whether a benefit is observed and the magnitude of the benefit) is more complex and depends on factors such as the degree of hearing loss and, potentially, age [e.g., Gallun ]. In the Kidd study, only the listeners with the poorest performance under “natural” spatial hearing conditions (i.e., simulated by KEMAR head-related transfer functions) obtained a significant benefit whereas all of the cochlear implant subjects in Yun obtained a significant benefit from the beamformer. In some cases, NH listeners achieved significantly better SRTs without the beamformer using natural binaural cues in unaided listening (Kidd ). These results also suggested that person-to-person performance varied substantially within both groups—a finding that is typical for high-IM conditions [e.g., Kidd and Colburn (2017)].

Prior to the current project, little was known about the extent to which beamforming might provide a benefit to PWA listeners. Importantly, PWA differ from populations in which the beamformer has previously been tested, in that they exhibit known cognitive-linguistic deficits that are central in origin and, as a result, the challenges they encounter in cocktail party listening situations could be rooted in somewhat different factors. Additionally, PWA tend to be older than many of the listeners in which beamforming has previously been tested, introducing the possibility of age-related cognitive differences, as well as possible differences related to age-related hearing loss. Given these differences, directly measuring the effect of beamforming in PWA listeners will provide valuable information about whether it could be an appropriate compensatory aid for use in this population.

Additionally, assessing the benefit of acoustic beamforming in PWA may provide information about the factors that facilitate or hinder the ability of PWA to understand speech in complex acoustic environments. While our previous work has found that PWA perform more poorly than controls under a condition where maskers consist of intelligible speech spatially separated from the target (Villard and Kidd, 2019), the precise reason(s) for this are not yet known. Because beamforming provides the listener with both a distinct advantage (an improved TMR) and a distinct disadvantage (removal of binaural spatial cues), relative to a natural listening condition, it may allow us to learn more about what drives PWA performance. For example, if the reason PWA perform more poorly under naturalistic listening conditions involving binaural spatial cues is simply that they have difficulty taking advantage of those cues to separate speech streams, then beamforming—which removes these cues and replaces them with an improved TMR delivered through a single channel—might be expected to provide a notable benefit for PWA. However, if the poorer performance observed in PWA arises not from difficulty utilizing binaural spatial cues but rather from difficulty with higher level cognitive skills involved in disentangling intelligible speech streams, then beamforming might offer somewhat less of an advantage in masked speech recognition (although, it should be noted, an improved TMR could also result in somewhat of a reduced cognitive processing load for the listener).

Modifications of psychoacoustic methods for listeners with aphasia

The process of assessing the benefit of beamforming in PWA listeners is complicated by the fact that the methodology used in our previous work on beamforming—and indeed, in many typical psychoacoustic experiments or standard clinical procedures (e.g., verbal word recognition/repetition tasks) assessing speech intelligibility—may not be appropriate for testing PWA (Zhang ). Because addressing questions related to speech intelligibility typically necessitates extensive measurements (i.e., many repetitions across numerous conditions), the psychoacoustic procedure of choice often involves the use of closed-set, forced-choice, matrix-style speech identification methods designed to minimize the effects of learning on performance. Such methods afford the advantage that the limited set of items can be vetted in advance to assure familiarity to the participant and, because the items are selected at random on each trial from a closed set, the concerns about prior exposure to the specific test items (e.g., a limited number of complete sentences available, limited numbers of talkers available, etc.) that may confound multiple repetitions of lists of open set materials are moot [e.g., Webster (1983)]. In such procedures, performance typically is evaluated by presenting auditory stimuli throughout a sequence of trials, with a response made after each stimulus/trial often via a graphical user interface (GUI) by mouse-clicking or touching the response alternative on the computer screen. The GUI usually contains a list, or lists, of words from which the listener is asked to select the target words one by one. For example, a typical target sentence might contain five (or more) words, and the listener might be presented with a large GUI containing five (or more) lists of response options, one for each word in the sentence (or, alternatively, a series of GUIs that appear on the screen one at a time, one for each word in the target sentence).

While these matrix-style experiments offer a number of important advantages, they may cause unintended experimental confounds if administered to listeners with known cognitive-linguistic impairments. To begin with, while these tests are intended to measure the listener's speech recognition abilities, the response selection process presents a number of additional demands, as listeners must efficiently read and/or scan through multiple lists of response options while continuing to hold the target sentence in memory. Such demands are not thought to be particularly taxing for listeners in the general population. However, because many PWA exhibit deficits in reading, scanning, and working memory, these response-related demands could present a substantial additional challenge, resulting in artificially reduced performance for PWA despite adequate recognition of target speech. Additionally, many PWA have impaired verbal repetition skills, which may make it difficult or impossible for them to use common strategies such as verbal rehearsal to assist them with their responses. Therefore, the development of approaches to control/minimize these confounds so that psychoacoustic measures may be used to obtain valid measurements of speech intelligibility in PWA listeners is a key element of assessment.

Aims of the current study

The current study had three aims. The first and primary aim was simply to determine whether acoustic beamforming could provide a benefit for PWA in understanding speech in acoustically complex environments. Although prior work had demonstrated that beamforming can provide a benefit for listeners with SNHL, it could not be assumed that PWA would receive a similar benefit, despite the evidence suggesting that persons with SNHL and PWA both experience difficulty understanding speech in complex acoustic environments [e.g., Kidd and Villard and Kidd (2019)]. This is because the factors underlying these two groups' difficulties almost certainly differ, particularly in terms of whether the limitation on performance is predominantly peripheral or central in origin. For listeners with SNHL, the poor performance for spatially separated speech and masker sources likely is due to a degraded peripheral representation of the sounds which is known to increase EM [e.g., Arbogast , Marrone , and Best ], though for older listeners with SNHL, cognitive factors could also be at play (Gallun ). The single channel beamformer eliminates the perception of spatial separation of sources that occurs through normal binaural hearing and therefore eliminates the benefits of using interaural differences to enhance source segregation. Balanced against that loss of spatial perception is the increase in signal-to-noise ratio (S/N) from the spatial tuning of the beamformer. In order to solve the source segregation problem using the beamformer in a multiple talker sound field, the listener must rely on the improvement in relative level of the target source, as well as the different voice characteristics, to disentangle the talkers. Because PWA presumably do not have the same peripheral deficit as SNHL (e.g., reduced frequency and time resolution due to sensorineural pathology), the improvement in S/N from the beamformer may not compensate for the loss of the percept of spatial segregation of sounds to a similar degree and thus may not provide the same benefit for PWA as for SNHL, or may do so under some masking conditions and not others (e.g., high EM vs high IM). Notably, because aphasia is more common in older individuals, many PWA may also have some age-related peripheral hearing loss, and thus may also experience increased EM. However, the limited evidence available has shown that additional, centrally based processing problems likely are present (Villard and Kidd, 2019), potentially resulting in a mixed peripheral-central processing deficit. As discussed earlier, the extent to which PWA can utilize various sound source segregation cues currently is not known and the extent to which enhancing specific cues—such as segregation of competing speech sounds by level from a single-channel beamformer—is beneficial also is not clear. This study therefore examined the extent to which acoustic beamforming could improve speech recognition in PWA.

The second aim of the study was to examine the effect of acoustic beamforming as a front-end signal processing strategy on masked speech recognition in a group of controls who were age- and hearing-matched to the PWA listeners, and to compare this effect to that seen in the PWA group. Because PWA and age- and hearing-matched (i.e., audiometrically similar) controls would be assumed to have similar peripheral hearing abilities, but to differ in central processing abilities, a comparison of the effect of beamforming on these two groups could help to clarify aspects of the central processing difficulties observed in PWA. Thus, the potential benefit of the beamformer was examined under both high-EM, low-IM conditions and high-IM, low-EM conditions, in both PWA and controls. Our expectation was that differences between the PWA and control listeners, if present, would be more apparent in a high-IM speech-on-speech masking condition than a high-EM speech-on-noise masking condition because aphasia is a central nervous system disorder.

The third and final aim of the study was to assess the feasibility of modifications of standard psychoacoustic/speech recognition methods for use with PWA. As discussed above, typical psychoacoustic methods using matrix-style sentences that require participants to read through word lists in order to respond may present challenges/confounds for PWA participants. In our previous study, we sought to bypass the majority of these confounds by using a very small response set containing only highly imageable nouns, with pictures as response options instead of written words (Villard and Kidd, 2019). Although this approach was effective for the purposes of that study, it did have some limitations, particularly with respect to the types of words that could easily be represented by graphical images. The current study, therefore, took a different approach to adapting the demands of the task to the abilities of the participant. Here we retained the speech matrix test used previously in studies of the benefits of beamforming for NH and SNHL participants (Kidd ), a test that depends on the use of written words as response options. However, modifications of sentence length and the number of available response items in each syntactic category were made to accommodate the abilities of individual PWA. Because the PWA population is quite heterogeneous, with different individuals displaying different degrees of difficulty with tasks such as reading and working memory, we chose to determine the extent of these modifications individually for each participant, using a combination of rule-based decision-making and clinical judgment, as outlined in Sec. II. The goal of this approach was to employ stimuli that were closer to those used in previous psychoacoustic experiments examining the effects of acoustic beamforming, while taking into consideration each PWA listener's specific limitations. Importantly, this effort has implications not only for assessing the benefit of acoustic beamforming in PWA but potentially also for the investigation of speech recognition in complex listening conditions in other populations with impaired language and/or cognition who cannot reliably be tested using standard methods.

METHODS

Participants

A total of ten listeners served as participants in this experiment. An eleventh participant was dismissed after failing to meet the minimum performance criteria for participation (as explained further below). Of the ten remaining participants, five demonstrated aphasia resulting from a stroke in the language-dominant hemisphere (as did the eleventh participant). All participants with aphasia were in the chronic stage of recovery, meaning that their stroke had occurred more than 12 months prior to participation. Participants were recruited through existing participant databases at Boston University and through online advertisements. All participants demonstrated visual acuity that was adequate for task completion.

Each PWA participant's aphasia type and aphasia severity were identified using Part 1 of the Western Aphasia Battery-Revised (WAB-R) (Kertesz, 2007), a standardized language measure. WAB-R results indicated that two participants (PWA1 and PWA6) exhibited Broca's aphasia, a non-fluent aphasia type characterized by notable difficulty with word-finding and sentence formulation. The remaining four (PWA2, PWA3, PWA4, and PWA5) exhibited anomic aphasia, a fluent aphasia type characterized primarily by milder expressive word-finding difficulty. The WAB-R also provides Aphasia Quotients (AQs) indicative of overall aphasia severity. These scores suggest that PWA1 and PWA6 each exhibited a moderate aphasia, while the remaining four PWA participants exhibited a mild aphasia. In order to collect information about participants' selective attention abilities, the Map Search and Elevator Counting with Distraction subtests of the Test of Everyday Attention (TEA) (Robertson ) were also administered. The Map Search task requires participants to quickly locate as many instances as possible of a specific visual symbol on a visually cluttered map, and the Elevator Counting with Distraction task requires participants to attend to, count, and report the number of target tones heard while ignoring non-target tones (participants were offered the use of a number line during each response period to point to their answer rather than verbalizing it, if they preferred). Please see Table I for information on standardized test results in PWA participants, as well as possible score ranges.

TABLE I.

Standardized testing results for PWA participants. AQ range: 0–100; lower score indicates a greater deficit. An AQ of 51–75 is considered to indicate moderate aphasia; an AQ of 76 and above is considered to indicate mild aphasia. Map search range: 0–80; lower score indicates poorer performance. Elevator Counting w/Distraction task range: 0–10; lower score indicates poorer performance.

	Aphasia type	WAB AQ	TEA: Map search (2 min)	TEA: Elevator counting w/ distraction
PWA1	Broca's	63	74	9
PWA2	Anomic	96	44	9
PWA3	Anomic	96	56	1
PWA4	Anomic	98	35	2
PWA5	Anomic	90	23	6
PWA6	Broca's	59	21	2

The remaining five participants reported no history of stroke, brain injury, or other neurological event/disease and served as age- and hearing-matched controls. Each control participant was matched with one PWA participant, resulting in five pairs of participants. The age difference between the PWA participant and control participant within a given matched pair was no more than three years. Efforts were also made to match pairs according to hearing profile; however, matching by age was prioritized. Because of the challenge of matching across two parameters, one pair (PWA5 and C5) had somewhat mismatched hearing profiles. Please see Table II for details on the five matched pairs.

TABLE II.

Information on matched PWA-control pairs. 4 F-PTA = four-frequency pure tone average, or the average of pure tone thresholds at 500 Hz, 1 kHz, 2 kHz, and 4 kHz.

	Sex	Age	4F-PTA (left ear)	4F-PTA (right ear)
PWA1	M	54	15.0	12.5
PWA2	M	53	25.0	17.5
PWA3	F	61	9.4	8.8
PWA4	F	56	15.6	13.8
PWA5	M	67	32.5	32.5
	Average:	58.2	19.5	17.0
C1	F	56	11.0	16.3
C2	F	56	21.0	28.8
C3	M	62	9.4	8.8
C4	M	56	6.9	5.0
C5	M	64	11.3	8.8
	Average:	58.8	11.9	13.5

All participants completed pure tone hearing testing in each ear, and all participants demonstrated some degree of hearing loss [see Figs. 1(a) and 1(b) for average audiograms for PWA and controls]. This loss of sensitivity was generally greater at higher frequencies and was believed (based on participant report) to have been acquired in adulthood. No participant with aphasia reported any perceived link between their stroke history and hearing sensitivity. Hearing loss was relatively mild across participants, and no participants reported current or past use of hearing aids. This study was overseen by the Institutional Review Board at Boston University.

FIG. 1.

(a) Average pure tone audiograms for PWA. (b) Average pure tone audiograms for controls.

Experimental stimuli

Auditory stimuli consisted of recordings of 40 single words drawn from an 8 × 5 matrix (8 names, 8 verbs, 8 numbers, 8 adjectives, and 8 objects; see Table III) that has been used in a number of previous psychoacoustic experiments involving speech masking (Kidd ; Swaminathan ; Clayton ), including experiments on beamforming (Kidd ). Eight different recordings of each word were used in the study, each one spoken by a different female talker (i.e., each of eight talkers recorded the entire set of words), for a total of 320 total single-word recordings. Visual stimuli consisted of columns of typed words presented on a GUI on a computer screen. Stimuli were presented, and data were collected, using custom software in matlab (MathWorks, Inc., Natick, MA).

TABLE III.

Full experimental matrix.

Names	Verbs	Numbers	Adjectives	Objects
Bob	bought	two	big	bags
Jane	found	three	cheap	cards
Jill	gave	four	green	gloves
Lynn	held	five	hot	hats
Mike	lost	six	new	pens
Pat	saw	eight	old	shoes
Sam	sold	nine	red	socks
Sue	took	ten	small	toys

Individualized modifications and frequency-specific gain

In previous experiments in our laboratory, participants typically have been presented with auditory signals consisting of sentences having the structure