Earinter: A Closed-Loop System for Eating Pace Regulation with Just-in-Time Intervention Using Commodity Earbuds

TL;DR

Earinter presents a closed-loop system that uses commodity earbuds to sense chewing via a bone-conduction sensor, estimate eating pace on-device as chews per swallow (CPS), and deliver just-in-time auditory interventions during daily meals. The sensing pipeline achieves robust chewing detection ($F1=0.97$) and accurate CPS estimation ($MAE=0.18\pm0.13$ chews/min; $MAE=3.65\pm3.86$ chews/swallow), enabling reliable real-time regulation. Grounded in Dual Systems Theory, Earinter employs theory-based pre-, in-, and post-meal prompts, refined through Wizard-of-Oz pilots, to form a practical JIT intervention within a single wearable device. In a 13-day within-subject field study (N=14), Earinter significantly increased CPS and reduced eating speed, with evidence of short-term carryover and acceptable user burden and usability. The work demonstrates the feasibility and impact of a single-earbud closed-loop system for dynamic ingestion regulation in real-world settings, highlighting the potential of earables as practical platforms for JIT-driven habit modification.

Abstract

Rapid eating is common yet difficult to regulate in situ, partly because people seldom notice pace changes and sustained self-monitoring is effortful. We present Earinter, a commodity-earbud-based closed-loop system that integrates in-the-wild sensing, real-time reasoning, and theory-grounded just-in-time (JIT) intervention to regulate eating pace during daily meals. Earinter repurposes the earbud's bone-conduction voice sensor to capture chewing-related vibrations and estimate eating pace as chews per swallow (CPS) for on-device inference. With data collected equally across in-lab and in-the-wild sessions, Earinter achieves reliable chewing detection (F1 = 0.97) and accurate eating pace estimation (MAE: 0.18 $\pm$ 0.13 chews/min, 3.65 $\pm$ 3.86 chews/swallow), enabling robust tracking for closed-loop use. Guided by Dual Systems Theory and refined through two Wizard-of-Oz pilots, Earinter adopts a user-friendly design for JIT intervention content and delivery policy in daily meals. In a 13-day within-subject field study (N=14), the closed-loop system significantly increased CPS and reduced food-consumption speed, with statistical signs of carryover on retention-probe days and acceptable user burden. Our findings highlight how single-modality commodity earables can support practical, theory-driven closed-loop JIT interventions for regulating eating pace in the wild.

Figures (8)

• Figure 2: Closed-loop system overview of Earinter for eating-pace regulation. (a) Sensing and reasoning. The commodity earbud captures body-conducted vibrations during chewing and swallow. Candidate chewing cycles are segments from the signal, and then a lightweight model detect specific action. Swallow is inferred with a heuristic method. (b) Intervention. Eating pace estimation drives just-in-time intervention grounded in Dual Systems Theory, including pre-meal goal activation, in-meal prompts informed by specific theories, and post-meal reflection. Interventions follow the mechanism with a predefined threshold, cooldown intervals and diversified prompt lengths to reduce user burden.
• Figure 3: Data collection platform and experimental protocol. Eating behaviors were recorded across two environments (lab and cafeteria) using: (A) an eye-level iPhone for video ground truth, (B) integrated bone-conduction sensors in Honor Earbuds 3 Pro for capturing ear-canal vibrations, and (C) a laryngophone serving as a high-fidelity reference for precise chewing and swallowing annotation.
• Figure 4: Swallow detection foundation and the eating pace sensing pipeline. (A) Probability density distribution of durations between consecutive chewing cycles. The distinct temporal separation between intervals with swallows ($\mu = 1.919s$) and without swallows ($\mu = 0.424s$) validates the interval-inspired hypothesis for swallow detection. (B) The proposed sensing pipeline, which integrates candicate chewing segmentation, deep learning-based chewing confirmation, interval-inspired swallow detection to estimate the eating pace.
• Figure 5: User study overview for Condition $\rightarrow$ Experiment. (a) Timeline of the 13-day within-subject field study: a baseline day (no intervention), followed by two 6-day phases with five active days under each condition (Control: only pre-meal reminder; Experiment: pre-meal reminder & in-meal JIT intervention & post-meal feedback) and a retention day with no support. Dots and triangles mark questionnaire and interview. (b) Daily procedure for data collecting: participants measured body weight and fat, measured food weight before eating, ate with earbuds, and weighed leftovers to estimate intake.
• Figure 6: Representative eating-pace signals for P5 from the second day in Experiment condition. The top plot shows time-based cumulative chewing activity, and the bottom shows chews per swallow. Star markers indicate moments when Earinter triggered JIT prompts based on the real-time estimated pace.