Predicting and measuring decision rules for social recognition in a Neotropical frog

Many animals use signals, such as vocalizations, to recognize familiar individuals. However, animals risk making recognition mistakes because the signal properties of different individuals often overlap due to within-individual variation in signal production. To understand the relationship between signal variation and decision rules for social recognition, we studied male golden rocket frogs, which recognize the calls of territory neighbors and respond less aggressively to a neighbor's calls than to the calls of strangers. We quantified patterns of individual variation in acoustic properties of calls and predicted optimal discrimination thresholds using a signal detection theory model of receiver utility that incorporated signal variation, the payoffs of correct and incorrect decisions, and the rates of encounters with neighbors and strangers. We then experimentally determined thresholds for discriminating between neighbors and strangers using a habituation-discrimination experiment with territorial males in the field. Males required a threshold difference between 9% and 12% to discriminate between calls differing in temporal properties; this threshold matched those predicted by a signal detection theory model under ecologically realistic assumptions of infrequent encounters with strangers and relatively costly missed detections of strangers. We demonstrate empirically that receivers group continuous variation in vocalizations into discrete social categories and show that signal detection theory can be applied to investigate evolved decision rules.

BlinkListener

Eye blink detection plays a key role in many real-life applications such as Human-Computer Interaction (HCI), drowsy driving prevention and eye disease detection. Although traditional camera-based techniques are promising, multiple issues hinder their wide adoption including the privacy concern, strict lighting condition and line-of-sight (LoS) requirements. On the other hand, wireless sensing without a need for dedicated sensors gains a tremendous amount of attention in recent years. Among the wireless signals utilized for sensing, acoustic signals show a unique potential for fine-grained sensing owing to their low propagation speed in the air. Another trend favoring acoustic sensing is the wide availability of speakers and microphones in commodity devices. Promising progress has been achieved in fine-grained human motion sensing such as breathing using acoustic signals. However, it is still very challenging to employ acoustic signals for eye blink detection due to the unique characteristics of eye blink (i.e., subtle, sparse and aperiodic) and severe interference (i.e., from the human target himself and surrounding objects). We find that even the very subtle involuntary head movement induced by breathing can severely interfere with eye blink detection. In this work, for the first time, we propose a system called BlinkListener to sense the subtle eye blink motion using acoustic signals in a contact-free manner. We first quantitatively model the relationship between signal variation and the subtle movements caused by eye blink and interference. Then, we propose a novel method that exploits the "harmful" interference to maximize the subtle signal variation induced by eye blinks. We implement BlinkListener on both a research-purpose platform (Bela) and a commodity smartphone (iPhone 5c). Experiment results show that BlinkListener can achieve robust performance with a median detection accuracy of 95%. Our system can achieve high accuracies when the smartphone is held in hand, the target wears glasses/sunglasses and in the presence of strong interference with people moving around.