Differential privacy concepts have been successfully used to protect anonymity of individuals in population-scale analysis. Sharing of mobile sensor data, especially physiological data, raise different privacy challenges, that of protecting private behaviors that can be revealed from time series of sensor data. Existing privacy mechanisms rely on noise addition and data perturbation. But the accuracy requirement on inferences drawn from physiological data, together with well-established limits within which these data values occur, render traditional privacy mechanisms inapplicable. In this work, we define a new behavioral privacy metric based on differential privacy and propose a novel data substitution mechanism to protect behavioral privacy. We evaluate the efficacy of our scheme using 660 hours of ECG, respiration, and activity data collected from 43 participants and demonstrate that it is possible to retain meaningful utility, in terms of inference accuracy (90%), while simultaneously preserving the privacy of sensitive behaviors.

Considerable work suggests the dominant syllable rhythm of the acoustic envelope is remarkably similar across languages (∼4–5 Hz) and that oscillatory brain activity tracks these quasiperiodic rhythms to facilitate speech processing. However, whether this fundamental periodicity represents a common organizing principle in both auditory and motor systems involved in speech has not been explicitly tested. To evaluate relations between entrainment in the perceptual and production domains, we measured individuals’ (i) neuroacoustic tracking of the EEG to speech trains and their (ii) simultaneous and non-simultaneous productions synchronized to syllable rates between 2.5 and 8.5 Hz. Productions made without concurrent auditory presentation isolated motor speech functions more purely. We show that neural synchronization flexibly adapts to the heard stimuli in a rate-dependent manner, but that phase locking is boosted near ∼4.5 Hz, the purported dominant rate of speech. Cued speech productions (recruit sensorimotor interaction) were optimal between 2.5 and 4.5 Hz, suggesting a low-frequency constraint on motor output and/or sensorimotor integration. In contrast, “pure” motor productions (without concurrent sound cues) were most precisely generated at rates of 4.5 and 5.5 Hz, paralleling the neuroacoustic data. Correlations further revealed strong links between receptive (EEG) and production synchronization abilities; individuals with stronger auditory-perceptual entrainment better matched speech rhythms motorically. Together, our findings support an intimate link between exogenous and endogenous rhythmic processing that is optimized at 4–5 Hz in both auditory and motor systems. Parallels across modalities could result from dynamics of the speech motor system coupled with experience-dependent tuning of the perceptual system via the sensorimotor interface.

Monitoring of in-person conversations has largely been done using acoustic sensors. In this paper, we propose a new method to detect moment-by-moment conversation episodes by analyzing breathing patterns captured by a mobile respiration sensor. Since breathing is affected by physical and cognitive activities, we develop a comprehensive method for cleaning, screening, and analyzing noisy respiration data captured in the field environment at individual breath cycle level. Using training data collected from a speech dynamics lab study with 12 participants, we show that our algorithm can identify each respiration cycle with 96.34% accuracy even in presence of walking. We present a Conditional Random Field, Context-Free Grammar (CRF-CFG) based conversation model, called rConverse, to classify respiration cycles into speech or non-speech, and subsequently infer conversation episodes. Our model achieves 82.7% accuracy for speech/non-speech classification and it identifies conversation episodes with 95.9% accuracy on lab data using a leave-one-subject-out cross-validation. Finally, the system is validated against audio ground-truth in a field study with 32 participants. rConverse identifies conversation episodes with 71.7% accuracy on 254 hours of field data. For comparison, the accuracy from a high-quality audio-recorder on the same data is 71.9%.

Surrounding context influences speech listening, resulting in dynamic shifts to category percepts. To examine its neural basis, event-related potentials (ERPs) were recorded during vowel identification with continua presented in random, forward, and backward orders to induce perceptual warping. Behaviorally, sequential order shifted individual listeners’ categorical boundary, versus random delivery, revealing perceptual warping (biasing) of the heard phonetic category dependent on recent stimulus history. ERPs revealed later (∼300 ms) activity localized to superior temporal and middle/inferior frontal gyri that predicted listeners’ hysteresis/enhanced contrast magnitudes. Findings demonstrate that interactions between frontotemporal brain regions govern top-down, stimulus history effects on speech categorization.