Gradient boosted decision trees reveal nuances of auditory discrimination behavior

Abstract

ABSTRACTPitch is a fundamental feature of a person’s voice, and a hallmark of human voice processing is recognizing a word regardless of voice pitch. Pitch continuity substantially improves our ability to hear a person’s voice in noise. However, it is not clear whether pitch continuity is uniquely human or whether it can be considered a more general feature of the mammalian auditory system. To assess this, 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. While ferrets were able to 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. While our models demonstrated that false alarms are primarily driven by non-acoustic factors, through our novel application of well-established machine learning algorithms, we were able to identify a subset of words that animals consistently confused with the target word. Overall, gradient-boosted regression and decision trees allowed us to tease apart acoustic and behavioral factors that shaped sound discrimination performance in this task, demonstrating that ferrets can identify complex sounds across variations in pitch but that they, like humans, utilize the F0 of sounds as a streaming cue.AUTHOR SUMMARYHearing is one of the aspects of life that we often take for granted until it is too late. According to the World Health Organization, by 2050, 1 in 10 people will have life-changing hearing loss. However, our understanding of how even normal brain mechanisms support hearing remains limited. This remains challenging because such work relies upon animal models, and most behavioral tasks implemented in animals remain impoverished versions of everyday challenges humans face when listening to speech in the presence of other competing sounds. Here, we used a novel behavioral paradigm designed to include this complexity and present a novel implementation of machine learning algorithms to understand how trained ferrets perform this task. Gradient-boosted regression and decision trees are well-established machine learning methods that do not require users to predetermine interaction effects. Through this machine learning method, we find that ferrets can perform the task across variations in voice pitch and benefit from continuity of pitch within a sound sequence to make their decisions. Our results suggest that this machine learning approach can analyze behavioral data in animal models and that ferrets can process complex aspects of sound stimuli to make choices under changing sound environments similar to humans.