Torsten Dau

When asked to discriminate between two stimuli that differ along some dimension, such as intensity, frequency, or orientation, humans exhibit a smallest stimulus difference, or threshold, below which discrimination becomes impossible. These limits on perceptual sensitivity have traditionally been explained by internal noise. Contemporary theories instead link perceptual sensitivity to efficient coding of the natural environment but have not been comprehensively tested. Here, we measured the psychoacoustic sensitivity of a computational model optimized for everyday sound localization and recognition tasks. We used linear classifiers operating on the model’s learned representations to estimate 1632 thresholds spanning the perceptual dimensions of sound. These model-derived thresholds closely matched thresholds measured behaviorally in humans on the same test battery. The results show that many classical psychophysical phenomena emerge from representations optimized for natural behavior and suggest that discrimination thresholds are determined by linear separability in these optimized representations, rather than intrinsic neural noise.

Hearing aid users commonly report listening-related fatigue, which may contribute to disengagement from demanding listening situations and negative psychosocial outcomes. Hearing rehabilitation currently lacks strategies to mitigate listening fatigue, partly because fatigue build-up and its consequences are difficult to measure beyond self-report. This study investigated whether value-based listening decisions and associated frontal cortical activity are modulated by fatigue. Thirty hearing aid users completed a one-hour listening paradigm while cortical hemodynamic activity was measured using functional near-infrared spectroscopy (fNIRS). On each trial, participants decided whether to engage in listening or rest based on an offer defined by expected task difficulty and reward, and rated their momentary fatigue. Fatigue ratings increased over time, and unsuccessful listening trials were associated with larger subsequent increases in fatigue, suggesting that performance contributed to changes in self-reported fatigue. Listen-versus-rest decisions followed the expected effort-reward trade-off, with participants more likely to accept offers that provided higher rewards and lower listening demands. Over the course of the experiment, participants became less likely to accept listening offers; however, neither time on task nor subjective fatigue altered the weighting of reward and difficulty. In contrast, frontal cortical activation during evaluation of accepted offers increased over time for high-difficulty offers with low or medium rewards. Follow-up analyses indicated that frontal oxygenation and listening decisions were more strongly associated with time on task than with subjective fatigue ratings. Together, these findings suggest that behavioral and neural measures obtained before listening capture fatigue-related processes beyond self-report, providing a basis for identifying and potentially addressing fatigue in hearing rehabilitation.

Vocoded speech is widely used to simulate aspects of cochlear implant listening, and modeling vocoded intelligibility in normal-hearing listeners can provide insight into auditory processing under electric hearing. Sentence recognition was measured using three masker types (speech-shaped noise, tone complex, and noise-modulated tone complex) and four processing conditions (unprocessed, bandpass-filtered, tone-vocoded, and tone-vocoded with simulated current spread). Masker type affected intelligibility in all conditions except vocoded speech with simulated current spread. A modified speech-based envelope power spectrum model with a correlation-based decision stage was fitted to the unprocessed speech-shaped-noise condition and then applied, without re-fitting, to the remaining conditions. The model captured masker-dependent intelligibility patterns for unprocessed, bandpass-filtered, and tone-vocoded speech without simulated current spread. Analyses of the model's internal representations suggested that vocoding-related intelligibility loss was driven mainly by the removal of higher-rate temporal-envelope modulations rather than by reduced spectral resolution. With simulated current spread, the model predicted the loss of benefit for unmodulated maskers but consistently overestimated intelligibility. These findings support the view that modulation masking is not the primary limitation in simulated electric hearing and point to envelope-energy and pitch-related distortions as important targets for future cochlear implant-oriented intelligibility models.

Despite significant advances in our understanding of human hearing and assistive hearing technologies, the benefits of hearing loss compensation continue to vary widely across individuals. A central challenge is the complexity and heterogeneity of hearing loss, whose perceptual consequences often extend far beyond reduced sensitivity. Conventional strategies rely on signal processing algorithms-such as spatial filtering, noise reduction, and dynamic range compression-to improve audibility and enhance target signals. These components are usually optimized in isolation, yet their combined effects may interfere with one another and therefore do not necessarily yield an overall benefit. More recently, machine learning techniques have been used to further improve the performance of individual components. To tailor compensation strategies to specific acoustic environments or listeners, various steering mechanisms have also been proposed, guided by acoustic cues, audiovisual input, or listener attention. While these approaches show promise, a consistent and objective computational target for optimization has yet to be established. As an alternative, auditory model-based strategies, increasingly combined with machine learning, have emerged. These approaches aim to minimize the discrepancy between simulated auditory representations of normal and impaired hearing, thereby providing a physiologically motivated optimization goal. Although both categories of strategies offer considerable potential, achieving effective compensation under real-time, real-world conditions remains a major challenge. This paper reviews opportunities and limitations of these approaches for individualized hearing aid compensation.

Traditional audiological tests often fail to replicate real-world listening environments, potentially leading to discrepancies between clinical assessments and patient experiences. Common tests involve sentences read aloud verbatim, which do not represent real-life speech and language. This study outlines the development of a more ecologically valid speech material and compares its psychometric properties to those of the more traditional, clear-speech Danish Sentence Test (DAST) corpus. Sentences were extracted from recordings of dialogues between a female talker and two partners, communicating in quiet and noisy conditions to elicit normal and raised vocal efforts. A total of 400 sentences were included in the Everyday Conversational Danish Sentence Test. These sentences were presented to 40 normal-hearing participants in stationary noise spectrally matched to each vocal effort, across five signal-to-noise ratios (SNRs) ranging from −10 to 0 dB in 2.5 dB steps. A psychometric function fitted across all sentences revealed a speech reception threshold (SRT) of −5.3 dB SNR and a slope of 12.1%/dB, similar to the psychometric properties of DAST (SRT: −5.17 dB SNR; slope: 11.5%/dB). Vocal effort significantly affected intelligibility, but the direction of the effect varied with SNR. Sentences were intelligibility-normalized and grouped into balanced lists for future implementations.

Objectives: The assessment of aided phoneme perception is essential in validating whether hearing aids and cochlear implants adequately meet the amplification and rehabilitation needs of children. Phonemes are commonly used in such assessments to minimize the influence of semantic context and vocabulary knowledge. Traditionally, they have been presented by clinicians using their own voices. This study directly compared the performance of phoneme detection, discrimination, and identification among children with normal hearing, children using hearing aids, children with cochlear implants, and adults with normal hearing. Laboratory procedures and recorded sound files were used throughout. In addition, phoneme discrimination performance was assessed in challenging acoustic environments—specifically those with noise and/or reverberation—mirroring the types of settings children frequently encounter. Design: A total of 60 children and 42 adults participated in the study. The children had an average age of 10 ± 2 yrs, and the adults an average age of 27 ± 5 yrs. The children were divided into 3 groups of 20: those with normal hearing, those using bilateral hearing aids, and those with bilateral cochlear implants. Children with hearing devices were recruited through the Copenhagen Hearing and Balance Center (Rigshospitalet, Denmark) and were tested using their own devices, all fitted according to present best practices. Aided phoneme detection and identification were assessed using the Ling 6(HL) CD sound files, while phoneme discrimination was evaluated using a subset of the DANOK speech corpus. Testing was conducted in a spatial hearing laboratory to examine the impact of noise and reverberation on performance. Results: The results of this study indicate that, aside from minor differences in low-frequency speech phonemes, there were no significant differences between children with hearing aids and those with cochlear implants in phoneme detection, identification, or discrimination across a range of acoustic conditions. All children demonstrated Ling sound detection thresholds in quiet within normal limits, with average thresholds across all Ling sounds below 20 dB HL for all groups (rising to a maximum average of 22 dB HL when analyzed per Ling sound and participant group). Phoneme discrimination was significantly affected by reverberation and noise in all children—more so than in adults—with the most pronounced effects observed in children using hearing aids or cochlear implants. Conclusions: The procedures for aided phoneme detection, discrimination, and identification presented in this study are valuable tools for objectively assessing outcomes and validating benefit in children using hearing aids and cochlear implants. This is particularly important for children with cochlear implants, as electroacoustic verification through real-ear measures is not possible for these devices. The results suggest that phoneme perception in children with hearing aids and those with cochlear implants—despite the fundamentally different nature of these technologies (acoustic versus electric stimulation)—does not differ significantly. The reduced performance in noisy and reverberant conditions among children, especially those with hearing devices, underscores the importance of optimizing signal to noise ratios and acoustic environments in children’s everyday settings to support speech perception, auditory development, and overall wellbeing.

Background: Evidence on the prevalence and nature of hearing impairments among patients with Alzheimer's disease and other types of dementia is limited and inconsistent. Objective: The objective of this study was to compare the proportion of individuals with hearing loss and the severity of hearing impairments in peripheral hearing and speech perception in noise of patients with mild cognitive impairment (MCI) and cognitively intact older adults. Methods: Participants included 40 patients with MCI recruited from a memory clinic and 60 community-dwelling cognitively intact older adults (Clinical Dementia Rating Score of 0 and Mini-Mental State Examination ≥ 26). All participants were assessed with pure-tone audiometry and a speech-in-noise test, and with cognitive tests tapping memory, processing speed, executive functions, attention, and language. Results: No significant difference was found in the proportion of individuals with peripheral hearing loss among patients with MCI and cognitively intact older adults (77.5% and 60%, respectively). Speech-in-noise did not differ between the groups either, not even when accounting for aided hearing in those participants that wore hearing aids in their daily life. About one third of patients with MCI did not perceive hearing problems, despite having objectively measured hearing loss. However, this proportion did not differ significantly from the group of cognitively intact older adults. Conclusions: Despite differences in cognitive status, this study did not show significant differences between the groups on measures of peripheral hearing, speech-in-noise, and subjective complaints of hearing when comparing memory clinic patients with cognitively intact older adults.

The development and evaluation of hearing rehabilitation strategies would greatly benefit from quantification of theoretical constructs related to communication success. Motivated by this, we present a model-based approach to analyze information exchange in a collaborative general knowledge decision-making task. Through a combination of experiments and simulations, we investigate how this model can be used to quantify the exchange of information between interlocutors. Experiments were conducted with ten triads (N = 30) to examine the impact of loud background noise on decision-making in collaborating triads. The group discussions took place in two different levels of background noise, 48dB and 78dB. An existing model of joint decision-making was extended to fit cases where decisions are made individually after engaging in a collaborative discussion. A maximum likelihood estimator for the model was derived and validated in terms of parameter recovery and sensitivity to participant response bias and was used to quantify the relative influence of group members on each other’s post-discussion decisions, formalized as a set of model weights. Four statistics were used to summarize the results: overall weight change, self-weighting, weight equality, and weight similarity. Background noise was found to significantly alter how participants influenced each other’s decisions, but the direction of change remained unclear. These findings demonstrate how group members’ influence on each other’s decisions can be quantified and suggest that loud background noise can have a tangible impact on how group decisions are formed.

Spectro-temporal modulation (STM) sensitivity has been proposed as a sensitive marker of speech intelligibility in challenging listening conditions, yet the underlying auditory mechanisms involved in STM detection remain incompletely understood. The present study measured STM detection thresholds in young normal-hearing and older hearing-impaired listeners and evaluated whether the revised Computational Auditory Signal Processing and Perception model (CASP) can account for individual performance. Thresholds were obtained for six modulation detection conditions, defined by combinations of spectral (0, 1, and 2 c/o) and temporal (4 and 12 Hz) rates. To individualize CASP, outer and inner hair cell loss estimates were obtained from audiometric and Adaptive Categorical Loudness Scaling (ACALOS) data. The results showed systematically elevated thresholds in older hearing-impaired listeners as compared to the young normal-hearing group, particularly at higher spectral rates. The model simulations reproduced overall threshold patterns, but substantially underestimated group differences and interindividual variability in the data. Moreover, the simulations showed limited sensitivity to estimates of outer and inner hair cell loss, supporting the idea that additional supra-threshold mechanisms contribute to STM deficits. While these findings demonstrate the potential of auditory models to predict STM performance, they also highlight the need for refined representations of peripheral and central processing to account for individual STM detection thresholds.