Non-speech body sounds such as sounds of food intake, breath, laughter, and cough contain invaluable information about our dietary behavior, respiratory physiology, and affect. With regard to food and beverage consumption, body sounds enable us to discriminate characteristics of food and drinks from the inherent qualities of those sounds. Longer term tracking of eating sound has a lot of potential in dietary monitoring applications. Breathing sound is generated by the friction caused by the air flow from our lungs through the vocal organs (e.g. trachea, larynx, etc.) to the mouth or nasal cavity. Breathing sounds are highly indicative of the condition of the lungs and other pulmonary processes. Similarly, some classes of non-speech body sounds (laughter, yawn etc.) can be highly indicative of affect.

The built-in microphones on smartphones have been used to capture louder body sounds (e.g. coughing, snoring, etc.), but in general, it is very challenging to exclusively use a smartphone for capturing and analyzing non-speech body sounds because many sounds generated by the human body are subtle and quiet. When using the built-in microphone on a smartphone, it is usually difficult to achieve a good microphone placement and even more difficult to maintain that placement. The quality of the recorded audio decreases significantly in the presence of different external sounds and ambient noise.

In this paper, we propose BodyBeat a novel mobile sensing framework for capturing and inferring a diverse range of body sounds in real-life scenarios. BodyBeat includes a custom-built microphone and a distributed computation framework that uses the ARM micro-controller of the custom microphone and an Android smartphone to optimize power, CPU, and memory usage. The custom-built microphone is designed to capture subtle body vibrations directly from the body surface without being perturbed by external sounds. The microphone is attached to a 3D printed neckpiece with a suspension mechanism to hinder any friction noise due to head movements. We examine the performance of BodyBeat and demonstrate that it outperforms other state-of-the-art microphones used for sensing non-speech body sounds.

Students Involved: Tauhidur Rahman