Gradient boosted decision trees reveal nuances of auditory discrimination behavior.

Animal psychophysics can generate rich behavioral datasets, often comprised of many 1000s of trials for an individual subject. Gradient-boosted models are a promising machine learning approach for analyzing such data, partly due to the tools that allow users to gain insight into how the model makes predictions. We trained ferrets to report a target word's presence, timing, and lateralization within a stream of consecutively presented non-target words. To assess the animals' ability to generalize across pitch, we manipulated the fundamental frequency (F0) of the speech stimuli across trials, and to assess the contribution of pitch to streaming, we roved the F0 from word token to token. We then implemented gradient-boosted regression and decision trees on the trial outcome and reaction time data to understand the behavioral factors behind the ferrets' decision-making. We visualized model contributions by implementing SHAPs feature importance and partial dependency plots. While ferrets could accurately perform the task across all pitch-shifted conditions, our models reveal subtle effects of shifting F0 on performance, with within-trial pitch shifting elevating false alarms and extending reaction times. Our models identified a subset of non-target words that animals commonly false alarmed to. Follow-up analysis demonstrated that the spectrotemporal similarity of target and non-target words rather than similarity in duration or amplitude waveform was the strongest predictor of the likelihood of false alarming. Finally, we compared the results with those obtained with traditional mixed effects models, revealing equivalent or better performance for the gradient-boosted models over these approaches.

Inferring the basis of binaural detection with a modified autoencoder.

The binaural system utilizes interaural timing cues to improve the detection of auditory signals presented in noise. In humans, the binaural mechanisms underlying this phenomenon cannot be directly measured and hence remain contentious. As an alternative, we trained modified autoencoder networks to mimic human-like behavior in a binaural detection task. The autoencoder architecture emphasizes interpretability and, hence, we "opened it up" to see if it could infer latent mechanisms underlying binaural detection. We found that the optimal networks automatically developed artificial neurons with sensitivity to timing cues and with dynamics consistent with a cross-correlation mechanism. These computations were similar to neural dynamics reported in animal models. That these computations emerged to account for human hearing attests to their generality as a solution for binaural signal detection. This study examines the utility of explanatory-driven neural network models and how they may be used to infer mechanisms of audition.

The role of temporal coherence and temporal predictability in the build-up of auditory grouping.

The cochlea decomposes sounds into separate frequency channels, from which the auditory brain must reconstruct the auditory scene. To do this the auditory system must make decisions about which frequency information should be grouped together, and which should remain distinct. Two key cues for grouping are temporal coherence, resulting from coherent changes in power across frequency, and temporal predictability, resulting from regular or predictable changes over time. To test how these cues contribute to the construction of a sound scene we present listeners with a range of precursor sounds, which act to prime the auditory system by providing information about each sounds structure, followed by a fixed masker in which participants were required to detect the presence of an embedded tone. By manipulating temporal coherence and/or temporal predictability in the precursor we assess how prior sound exposure influences subsequent auditory grouping. In Experiment 1, we measure the contribution of temporal predictability by presenting temporally regular or jittered precursors, and temporal coherence by using either narrow or broadband sounds, demonstrating that both independently contribute to masking/unmasking. In Experiment 2, we measure the relative impact of temporal coherence and temporal predictability and ask whether the influence of each in the precursor signifies an enhancement or interference of unmasking. We observed that interfering precursors produced the largest changes to thresholds.

Burden of rare variants in synaptic genes in patients with severe tinnitus: An exome based extreme phenotype study.

BACKGROUND: tinnitus is a heterogeneous condition associated with audiological and/or mental disorders. Chronic, severe tinnitus is reported in 1% of the population and it shows a relevant heritability, according to twins, adoptees and familial aggregation studies. The genetic contribution to severe tinnitus is unknown since large genomic studies include individuals with self-reported tinnitus and large heterogeneity in the phenotype. The aim of this study was to identify genes for severe tinnitus in patients with extreme phenotype. METHODS: for this extreme phenotype study, we used three different cohorts with European ancestry (Spanish with Meniere disease (MD), Swedes tinnitus and European generalized epilepsy). In addition, four independent control datasets were also used for comparisons. Whole-exome sequencing was performed for the MD and epilepsy cohorts and whole-genome sequencing was carried out in Swedes with tinnitus. FINDINGS: we found an enrichment of rare missense variants in 24 synaptic genes in a Spanish cohort, the most significant being PRUNE2, AKAP9, SORBS1, ITGAX, ANK2, KIF20B and TSC2 (p < 2E-04), when they were compared with reference datasets. This burden was replicated for ANK2 gene in a Swedish cohort with 97 tinnitus individuals, and in a subset of 34 Swedish patients with severe tinnitus for ANK2, AKAP9 and TSC2 genes (p < 2E-02). However, these associations were not significant in a third cohort of 701 generalized epilepsy individuals without tinnitus. Gene ontology (GO) and gene-set enrichment analyses revealed several pathways and biological processes involved in severe tinnitus, including membrane trafficking and cytoskeletal protein binding in neurons. INTERPRETATION: a burden of rare variants in ANK2, AKAP9 and TSC2 is associated with severe tinnitus. ANK2, encodes a cytoskeleton scaffolding protein that coordinates the assembly of several proteins, drives axonal branching and influences connectivity in neurons.

Seasonal weight changes in laboratory ferrets.

Ferrets (Mustela putorius furo) are a valuable animal model used in biomedical research. Like many animals, ferrets undergo significant variation in body weight seasonally, affected by photoperiod, and these variations complicate the use weight as an indicator of health status. To overcome this requires a better understanding of these seasonal weight changes. We provide a normative weight data set for the female ferret accounting for seasonal changes, and also investigate the effect of fluid regulation on weight change. Female ferrets (n = 39) underwent behavioural testing from May 2017 to August 2019 and were weighed daily, while housed in an animal care facility with controlled light exposure. In the winter (October to March), animals experienced 10 hours of light and 14 hours of dark, while in summer (March to October), this contingency was reversed. Individual animals varied in their body weight from approximately 700 to 1200 g. However, weights fluctuated with light cycle, with animals losing weight in summer, and gaining weight in winter such that they fluctuated between approximately 80% and 120% of their long-term average. Ferrets were weighed as part of their health assessment while experiencing water regulation for behavioural training. Water regulation superimposed additional weight changes on these seasonal fluctuations, with weight loss during the 5-day water regulation period being greater in summer than winter. Analysing the data with a Generalised Linear Model confirmed that the percentage decrease in weight per week was relatively constant throughout the summer months, while the percentage increase in body weight per week in winter decreased through the season. Finally, we noted that the timing of oestrus was reliably triggered by the increase in day length in spring. These data establish a normative benchmark for seasonal weight variation in female ferrets that can be incorporated into the health assessment of an animal's condition.

Mammalian behavior and physiology converge to confirm sharper cochlear tuning in humans.

Frequency analysis of sound by the cochlea is the most fundamental property of the auditory system. Despite its importance, the resolution of this frequency analysis in humans remains controversial. The controversy persists because the methods used to estimate tuning in humans are indirect and have not all been independently validated in other species. Some data suggest that human cochlear tuning is considerably sharper than that of laboratory animals, while others suggest little or no difference between species. We show here in a single species (ferret) that behavioral estimates of tuning bandwidths obtained using perceptual masking methods, and objective estimates obtained using otoacoustic emissions, both also employed in humans, agree closely with direct physiological measurements from single auditory-nerve fibers. Combined with human behavioral data, this outcome indicates that the frequency analysis performed by the human cochlea is of significantly higher resolution than found in common laboratory animals. This finding raises important questions about the evolutionary origins of human cochlear tuning, its role in the emergence of speech communication, and the mechanisms underlying our ability to separate and process natural sounds in complex acoustic environments.

ON-OFF receptive fields in auditory cortex diverge during development and contribute to directional sweep selectivity.

Neurons in the auditory cortex exhibit distinct frequency tuning to the onset and offset of sounds, but the cause and significance of ON and OFF receptive field (RF) organisation are not understood. Here we demonstrate that distinct ON and OFF frequency tuning is largely absent in immature mouse auditory cortex and is thus a consequence of cortical development. Simulations using a novel implementation of a standard Hebbian plasticity model show that the natural alternation of sound onset and offset is sufficient for the formation of non-overlapping adjacent ON and OFF RFs in cortical neurons. Our model predicts that ON/OFF RF arrangement contributes towards direction selectivity to frequency-modulated tone sweeps, which we confirm by neuronal recordings. These data reveal that a simple and universally accepted learning rule can explain the organisation of ON and OFF RFs and direction selectivity in the developing auditory cortex.

Spatial Processing Is Frequency Specific in Auditory Cortex But Not in the Midbrain.

The cochlea behaves like a bank of band-pass filters, segregating information into different frequency channels. Some aspects of perception reflect processing within individual channels, but others involve the integration of information across them. One instance of this is sound localization, which improves with increasing bandwidth. The processing of binaural cues for sound location has been studied extensively. However, although the advantage conferred by bandwidth is clear, we currently know little about how this additional information is combined to form our percept of space. We investigated the ability of cells in the auditory system of guinea pigs to compare interaural level differences (ILDs), a key localization cue, between tones of disparate frequencies in each ear. Cells in auditory cortex believed to be integral to ILD processing (excitatory from one ear, inhibitory from the other: EI cells) compare ILDs separately over restricted frequency ranges which are not consistent with their monaural tuning. In contrast, cells that are excitatory from both ears (EE cells) show no evidence of frequency-specific processing. Both cell types are explained by a model in which ILDs are computed within separate frequency channels and subsequently combined in a single cortical cell. Interestingly, ILD processing in all inferior colliculus cell types (EE and EI) is largely consistent with processing within single, matched-frequency channels from each ear. Our data suggest a clear constraint on the way that localization cues are integrated: cortical ILD tuning to broadband sounds is a composite of separate, frequency-specific, binaurally sensitive channels. This frequency-specific processing appears after the level of the midbrain.SIGNIFICANCE STATEMENT For some sensory modalities (e.g., somatosensation, vision), the spatial arrangement of the outside world is inherited by the brain from the periphery. The auditory periphery is arranged spatially by frequency, not spatial location. Therefore, our auditory perception of location must be synthesized from physical cues in separate frequency channels. There are multiple cues (e.g., timing, level, spectral cues), but even single cues (e.g., level differences) are frequency dependent. The synthesis of location must account for this frequency dependence, but it is not known how this might occur. Here, we investigated how interaural-level differences are combined across frequency along the ascending auditory system. We found that the integration in auditory cortex preserves the independence of the different-level cues in different frequency regions.

Comodulation Enhances Signal Detection via Priming of Auditory Cortical Circuits.

Acoustic environments are composed of complex overlapping sounds that the auditory system is required to segregate into discrete perceptual objects. The functions of distinct auditory processing stations in this challenging task are poorly understood. Here we show a direct role for mouse auditory cortex in detection and segregation of acoustic information. We measured the sensitivity of auditory cortical neurons to brief tones embedded in masking noise. By altering spectrotemporal characteristics of the masker, we reveal that sensitivity to pure tone stimuli is strongly enhanced in coherently modulated broadband noise, corresponding to the psychoacoustic phenomenon comodulation masking release. Improvements in detection were largest following priming periods of noise alone, indicating that cortical segregation is enhanced over time. Transient opsin-mediated silencing of auditory cortex during the priming period almost completely abolished these improvements, suggesting that cortical processing may play a direct and significant role in detection of quiet sounds in noisy environments. SIGNIFICANCE STATEMENT: Auditory systems are adept at detecting and segregating competing sound sources, but there is little direct evidence of how this process occurs in the mammalian auditory pathway. We demonstrate that coherent broadband noise enhances signal representation in auditory cortex, and that prolonged exposure to noise is necessary to produce this enhancement. Using optogenetic perturbation to selectively silence auditory cortex during early noise processing, we show that cortical processing plays a crucial role in the segregation of competing sounds.

Relating approach-to-target and detection tasks in animal psychoacoustics.

Psychophysical experiments seek to measure the limits of perception. While straightforward in humans, in animals they are time consuming. Choosing an appropriate task and interpreting measurements can be challenging. We investigated the localization of high-frequency auditory signals in noise using an "approach-to-target" task in ferrets, how task performance should be interpreted in terms of perception, and how the measurements relate to other types of tasks. To establish their general ability to localize, animals were first trained to discriminate broadband noise from 12 locations. Subsequently we tested their ability to discriminate between band-limited targets at 2 or 3 more widely spaced locations, in a continuous background noise. The ability to discriminate between 3 possible locations (-90°, 0°, 90°) of a 10-kHz pure tone decreased gradually over a wide range (>30 dB) of signal-to-noise ratios (SNRs). Location discrimination ability was better for wide band noise targets (0.5 and 2 octave). These results were consistent with localization ability limiting performance for pure tones. Discrimination of pure tones at 2 locations (-90/left, 90/right) was robust at positive SNRs, yielding psychometric functions which fell steeply at negative SNRs. Thresholds for discrimination were similar to previous tone-in-noise thresholds measured in ferrets using a yes/no task. Thus, using an approach-to-target task, sound "localization" in noise can reflect detectability or the ability to localize, depending on the stimulus configuration. Signal-detection-theory-based models were able to account for the results when discriminating between pure tones from 2- and 3-source locations. (PsycINFO Database Record

Behavioural estimates of auditory filter widths in ferrets using notched-noise maskers.

Frequency selectivity is a fundamental property of hearing which affects almost all aspects of auditory processing. Here auditory filter widths at 1, 3, 7, and 10 kHz were estimated from behavioural thresholds using the notched-noise method [Patterson, Nimmo-Smith, Weber, and Milroy, J. Acoust. Soc. Am. 72, 1788-1803 (1982)] in ferrets. The mean bandwidth was 21% of the signal frequency, excluding wider bandwidths at 1 kHz (65%). They were comparable although on average broader than equivalent measurements in other mammals (∼11%-20%), and wider than bandwidths measured from the auditory nerve in ferrets (∼18%). In non-human mammals there is considerable variation between individuals, species, and in the correspondence with auditory nerve tuning.