ABSTRACT. Auditory processing difficulties are often reported from individuals with neurodevelopmental disorders, including autism spectrum disorder. Changes in auditory brainstem functioning, including auditory brainstem responses (ABRs), precede other symptoms of autism spectrum disorder and thus may be a useful biomarker for early detection and intervention. However, the underlying biological mechanisms underlying these changes in ABRs in human infants are largely unknown. Prior research in the autistic brain, especially the cortex, has consistently reported alterations in excitatory-inhibitory balance as a potential biological mechanism underlying the disorder and alterations in auditory brainstem nuclei have also been reported.

We used a prenatal valproic acid (VPA) exposure model of autism in rats to examine changes in auditory brainstem responses at postnatal days 28-29. Waveform amplitudes and latencies were measured as well as the dB threshold for broadband click and tone stimuli (4-48kHz). In vitro patch clamp recordings were made in whole neurons in the superior olivary complex of the auditory brainstem from postnatal days 29-33. Intrinsic membrane properties, including rheobase, input resistance, and time constant were measured in current clamp mode. The morphology of the patched neurons was also assessed.

Current results suggest that prenatal VPA-exposed offspring display three different profiles of ABRs: (1) Elevated ABR thresholds at all tone stimuli; (2) Elevated ABR thresholds at high-frequency tones only (24-48 kHz); (3) No change in ABR thresholds. No changes in ABR thresholds with click stimuli were noted. At the single-cell level, we found an elevated rheobase and shorter time constant in principal neurons in the medial nucleus of the trapezoid body, but no major alterations in intrinsic membrane properties in principal neurons in the lateral superior olive. Morphological assessments of these neuron types are ongoing.

Our results suggest that there are individual differences between prenatal VPA-exposed offspring in ABRs and the ABRs are smaller overall, particularly with high-frequency tones. The finding that prenatal VPA exposure results in reduced intrinsic excitability of inhibitory neurons in the medial nucleus of the trapezoid body leads us to hypothesize that a deficit in excitatory-inhibitory balance in the lateral superior olive is an underlying component of altered ABRs. Deficits in the detection of speech patterns in challenging listening environments are well-known hallmarks of autism spectrum disorders in humans. The changes we have identified in the sound localization circuitry of the lateral superior olive of rats may provide mechanistic insight into these deficits, and help guide potential strategies for remediation.

ABSTRACT. Auditory evoked potentials (AEP) are objective diagnostic tools widely used in clinical settings and are increasingly valuable for assessing cochlear neural pathology. However, the complex relationship between AEP morphology and cochlear status complicates their interpretation. Computational models of AEPs based on cochlear function can elucidate the mechanisms underlying AEP generation mechanisms and help clinicians link individual differences in AEP morphology to specific cochlear pathologies.

We developed an AEP modeling framework that combines a state-of-the-art computational human auditory nerve (AN) model as the front-end with a convolution using a unitary response (UR) as the back-end. The stimulus-independent UR was fixed for specific electrode configurations: one set for peripheral potentials (tympanic membrane-to-mastoid) and another for brainstem potentials (mastoid-to-vertex). Simulated responses were validated against existing AEP datasets for transient stimuli (e.g., clicks) and pure and amplitude-modulated tones, representing frequency-following responses (FFR) and envelope-following responses (EFR) across a range of stimulus levels. Additionally, compound action potentials (CAP) and auditory brainstem responses (ABR) were recorded from 20 young participants with clinically normal thresholds in response to clicks and level-specific chirps (10 -80 dB nHL).

The model successfully replicated full AEP waveforms for both transient stimuli (CAPs, ABRs) and periodic stimuli (FFRs, EFRs) across various stimulus levels. It provided insights into AEP generation mechanisms within the cochlea, where the activation and interaction of distinct AN fibers explained key experimental observations, such as the ‘chirp’ benefit and stepwise amplitude growth functions. However, some discrepancies remained: the simulated CAP N1 and ABR wave V latency-level functions were too shallow, and amplitude growth functions plateaued at higher stimulus levels. These findings suggest that the current AN model may not fully account for the contributions and excitation spread of AN fibers at elevated stimulus levels.

Despite some limitations, this framework effectively simulated a wide range of AEPs across electrode configurations, capturing many stimulus- and level-dependent experimental observations in healthy peripheral systems. Electrocochleographic datasets can be used to further improve the predictive accuracy of human AN models. This modeling approach offers a systematic way to investigate how different cochlear pathologies, such as neural or outer hair cell loss, manifest in distinct AEP patterns. Additionally, large datasets of simulated AEPs from both healthy and impaired cochleae allow us to build machine-learning models to predict a patient’s exact cochlear damage profile from their AEPs measured in the clinics.

ABSTRACT. Future cochlear implants may use light instead of electricity to restore hearing in individuals with profound hearing loss. Unlike electrical stimulation, light can be precisely confined within cochlear fluid, potentially offering improved frequency selectivity. The anteroventral cochlear nucleus (AVCN) serves as the initial processing hub in the auditory pathway, integrating excitatory inputs from spiral ganglion neurons. The sound representation in AVCN spike trains is highly diverse, influenced by variations in cell types, morphology, connectivity, and molecular characteristics. This study investigates AVCN neural responses to optogenetic stimulation of the cochlea and explores the synaptic integration mechanisms driving these responses. Using Mongolian gerbils with spiral ganglion neurons optogenetically modified f-Chrimson, we activated the auditory pathway via cochlear optogenetic stimulation. A semi-stochastic stimulus paradigm, consisting of over 200 combinations of light pulses and inter-pulse intervals, was applied. Juxtacellular recordings from both spiral ganglion neurons and AVCN neurons were sequentially performed, complemented by simultaneous recordings of optically evoked compound action potentials to monitor cochlear activation. Our results provide new insights into how central auditory circuits process optogenetic inputs and highlight the synaptic mechanisms underlying these responses. These findings contribute to the broader understanding of central auditory processing and support the advancement of light-based cochlear implant technologies for hearing restoration.

ABSTRACT. Tinnitus has a high prevalence, and poor remission rate once present for more than four weeks. To understand the mechanisms and ‘purpose’ of tinnitus, it is crucial to understand the changes in neurological systems surrounding its onset, as well as persistence. This study featured a total of 27 Acute Tinnitus (time from onset mean 3.93 weeks, SD 2 weeks), 22 Chronic Tinnitus (>6 months from onset,) and 17 matched controls. 15 of the Acute Tinnitus group were also re-studied in the chronic stage. Based on EEG responses to 1kHz stimuli intensity-matched to individual audibility and tolerance, we calculated the Intensity Dependence of the Auditory Evoked Potential (IDAEP; increase in N1-P2 amplitude per dB increase in stimulus intensity), which inversely relates to serotonergic modulation. Results showed greater IDAEP in the acute tinnitus group, compared to the chronic tinnitus and control groups, and the same individuals re-studied later in the chronic tinnitus stage. Furthermore, the acute tinnitus group (but not chronic tinnitus and controls) showed a significant negative correlation between IDAEP and the dynamic range of stimuli used, suggesting relative preservation of reactivity to small acoustic changes and attenuation of reactivity to larger changes. Put another way, controls and chronic tinnitus showed larger N1-P2 differences between most and least intense stimuli with larger stimulus intensity differences, whilst tinnitus groups’ N1-P2 amplitude ranges were invariant of stimulus intensity range. We provisionally interpret these results as indicating that tinnitus initially occurs in the context of generalised auditory hyperreactivity, and acts as a specific type of high-level gain controller that stabilises central variance (i.e. maintains precision).

ABSTRACT. One of the main challenges in auditory perception is the detection and separation of target sounds from acoustic backgrounds, often referred to as the 'cocktail party problem'. This process involves multiple areas of the auditory cortex simultaneously, however, most imaging methods make a tradeoff between speed and resolution, making the large-scale activity inaccessible at high speed. Here we use the fast and very bright indicator jGCaMP8m together with high-speed imaging to achieve a temporal resolution of 10ms across the entire auditory cortex at indicator onset latencies of just ~5ms, including all secondary areas in normal hearing adult mice. Further, we combine widefield with 2p imaging to achieve both spatial coverage, single cell and depth resolution. We then study the representation of conspecific vocalizations embedded in acoustic texture backgrounds. The vocalizations are presented at random times and with random base frequencies in the context of a randomly drawn background texture. We then studied the neural representation of auditory textures and vocalizations, in particular comparing the vocalizations' silent representation with their embedded representation. We find that vocalizations and texture sounds activate overlapping but distinct cortical areas, however, vocalization responses show sustained responses in particular in secondary areas of the more laterally located temporal association area (TeA), which has already recently been highlighted for vocalization and predictive processing. This representation differs from their onset responses, which are more localized in primary areas. Adaptation to texture noise varied with multiple properties of the background texture, in particular cross-frequency correlations and temporal variance. This affected their spatial extent and the intensity of cortical responses. Additionally, extended exposure increased response areas exclusively for high-frequency vocalizations (32 kHz), indicating a frequency-dependent effect in neural representation. As most social vocalizations are in the ultrasonic range, this could reflect their ecological value. The cellular analysis is preliminary at this point, but points towards specific patterns of representation that depend on the subareas of the auditory cortex. Pupil and facial motion analysis indicate an arousal response upon presentation of both sound and vocalization. We conclude the high-speed imaging with fast and bright Calcium indicators provides an exciting toolset that is particularly relevant for auditory representations, where the time-scales are intrinsically fast and transient. The auditory cortex of the mouse can be studied encompassingly down to the cellular level, which opens a new level of access for understanding auditory cortical processing.

ABSTRACT. Individuals with sensorineural hearing loss differ widely in their ability to recognize speech in noise unrelated to their audiogram. Research has identified other contributing factors, both auditory and nonauditory, but these factors do not operate in isolation. In the present work a generic auditory model involving cue-sensitivity, cue-weighting, and decision noise, provided an integrative structural framework for identifying the relative contribution of auditory and nonauditory factors and their interaction.

Listeners were 12 normal-hearing (NH) and 13 hearing-impaired (HI) adults. Normal hearing was defined as pure-tone average thresholds ≤25dB HL from 0.25-4.0 kHz, and ≤30 dB HL up to 8.0 kHz. Listeners heard over headphones natural recordings of two talkers speaking sentences concurrently. Talker B (noise) was present on each trial, the other talker with equal probability was either A or C (targets) differing nominally from each other and B in fundamental frequency (F0) and azimuthal location (θ). The values of F0 and θ were perturbed slightly from trial to trial to simulate natural changes in speech prosody and different location of talkers speaking at different times. Listener tasks were target talker identification (speech separation) and target sentence recognition. For target talker identification three conditions involved changes in the acoustic, statistical, and linguistic properties of the talkers’ speech. To permit comparison across conditions, listener performance was expressed relative to that of an ideal observer, d'ideal, from signal detection theory (Green & Swets, 1966).

The results reveal the major source of intersubject variability for both NH and HI be a fixed effect across all conditions and tasks on the slope of the psychometric function relating listener performance d' to d'ideal. In the model, the fixed effect results from decision noise occurring late in the processing chain. Trial-by-trial analyses revealed a second much smaller contribution to the intersubject variability to be an interaction between cue sensitivity and cue weighting for some listeners. NH and HI group differences were greater for target sentence recognition than target talker identification due to the more subtle cues required for sentence recognition according to the model.

The results demonstrate the usefulness of an integrative approach to evaluating factors affecting speech recognition in noise. They suggest that while auditory factors largely distinguish speech recognition in noise for NH and HI groups, nonauditory factors associated with speech separation are largely responsible for the wide intersubject variation within groups. [work supported by NIH R01 DC001262-31].

ABSTRACT. The implementation of synthetic speech and automatic speech recognition for speech recognition testing could have a large impact on hearing healthcare, increasing both the efficiency of test development and testing while also decreasing variability due to tester differences (e.g., for clinicians who are hard-of-hearing, or second language learners of the tested language). Before such implementation occurs, we need to understand differences between synthetic and human speech, and we need to document the effectiveness of using automatic speech recognition for clinical and experimental utility.

The ability to recognize sentences in a speech-shaped noise was evaluated for synthetic and human targets. Experiment 1 used productions from five human talkers and from five synthetic talkers, generated using three different language models. Experiment 2 evaluated stimuli from human talkers and voice clones based on their speech. Automatic speech recognition was used, along with human scoring, to assess performance of the open-set responses. After speech recognition testing, participants provided perceptual ratings using a Likert scale to indicate whether the talker was human or synthetic. Recognition scores were evaluated by talker type (human, synthetic or clone), and as a function of the perceptual ratings. Automatic speech recognition scores were compared to those assigned by experimenters who were in the testing room with participants or listening to recordings of participant responses.

For Experiment 1, there was marked variability in recognition performance across talkers, with more inter-talker variability for humans than synthetic speech. Automatic speech recognition reliability was high and in agreement with human scoring (~98% of scores were the same between the two scoring methods). Perceptual ratings varied widely between talkers, and participants had difficulty confidently assessing which productions were human and which were synthetic. However, these perceptual ratings were not correlated with recognition scores. Data for Experiment 2 are currently being collected.

The use of synthetic speech and automatic speech recognition for open-set speech recognition testing has high promise for improving the delivery of speech recognition assessments and improving the inclusion of more languages in testing. Future work is needed to better understand the psychometric properties of synthetic speech before moving in this direction.

ABSTRACT. Speech-in-noise (SiN) recognition is challenging due to auditory and cognitive interference from competing sounds (i.e., maskers). Effective processing requires listeners to adapt strategies based on masker type to separate target speech from noise and integrate relevant information into meaningful representations. This task often presents distinct challenges for autistic individuals, who typically exhibit atypical auditory and cognitive profiles. This study examined neural and behavioural SiN processing in 31 autistic and 31 non-autistic adults with matched demographic and cognitive backgrounds. Auditory temporal processing was assessed using EEG-derived temporal response functions (TRF), while semantic integration was measured via the event-related potential N400 component during an auditory semantic congruency task across three background conditions: quiet, babble noise, and single-speaker speech noise. At the behavioural level, no group or group-related effects were observed, indicating that autistic participants performed as well as non-autistic participants. However, at the neural level, autistic participants exhibited reduced and delayed TRF and N400 responses compared to non-autistic participants, reflecting differences in neural tracking of auditory signals and semantic integration across all conditions. Moreover, non-autistic participants demonstrated a masker-specific trade-off between auditory and lexical-semantic processing resources, showing enhanced semantic integration but reduced neural tracking of auditory features when managing linguistic competition introduced by intelligible speech maskers. In contrast, the autistic group showed no modulation of neural responses based on masker type. Correlation analyses further suggested that autistic participants may draw on verbal and general cognitive abilities as compensatory strategies to support comparable behavioural performance. This study highlights the distinct strategies autistic individuals use to manage multi-level processing demands during SiN tasks. These findings provide novel insights into how atypical auditory processing interacts with cognitive mechanisms to shape auditory perception in autism. They also lay a foundation for developing models and interventions that respect and build upon these alternative strategies, addressing both auditory and cognitive aspects of SiN perception in autism.