Release from informational masking by auditory stream segregation: perception and its neural correlate

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.

Low-high-low or high-low-high? Pattern effects on sequential auditory scene analysis

Sequential auditory scene analysis (ASA) is often studied using sequences of two alternating tones, such as ABAB or ABA_, with "_" denoting a silent gap, and "A" and "B" sine tones differing in frequency (nominally low and high). Many studies implicitly assume that the specific arrangement (ABAB vs ABA_, as well as low-high-low vs high-low-high within ABA_) plays a negligible role, such that decisions about the tone pattern can be governed by other considerations. To explicitly test this assumption, a systematic comparison of different tone patterns for two-tone sequences was performed in three different experiments. Participants were asked to report whether they perceived the sequences as originating from a single sound source (integrated) or from two interleaved sources (segregated). Results indicate that core findings of sequential ASA, such as an effect of frequency separation on the proportion of integrated and segregated percepts, are similar across the different patterns during prolonged listening. However, at sequence onset, the integrated percept was more likely to be reported by the participants in ABA_low-high-low than in ABA_high-low-high sequences. This asymmetry is important for models of sequential ASA, since the formation of percepts at onset is an integral part of understanding how auditory interpretations build up.

Uncertainty-based informational masking in a vowel discrimination task for young and old Mongolian gerbils

Informational masking emerges with processing of complex sounds in the central auditory system and can be affected by uncertainty emerging from trial-to-trial variation of stimulus features. Uncertainty can be non-informative but confusing and thus mask otherwise salient stimulus changes resulting in increased discrimination thresholds. With increasing age, the ability for processing of such complex sound scenes degrades. Here, 6 young and 4 old gerbils were tested behaviorally in a vowel discrimination task. Animals were trained to discriminate between sequentially presented target and reference vowels of the vowel pair/I/-/i/. Reference and target vowels were generated shifting the three formants of the reference vowel in steps towards the formants of the target vowels. Non-informative but distracting uncertainty was introduced by random changes in location, level, fundamental frequency or all three features combined. Young gerbils tested with uncertainty for the target or target and reference vowels showed similar informational masking effects for both conditions. Young and old gerbils were tested with uncertainty for the target vowels only. Old gerbils showed no threshold increase discriminating vowels without uncertainty in comparison with young gerbils. Introducing uncertainty, vowel discrimination thresholds increased for young and old gerbils and vowel discrimination thresholds increased most when presenting all three uncertainty features combined. Old gerbils were more susceptible to non-informative uncertainty and their thresholds increased more than thresholds of young gerbils. Gerbils' vowel discrimination thresholds are compared to human performance in the same task (Eipert et al., 2019).