EchoMotion: enhancing exercise analysis with acoustic sensing

Mosuily, Mohammed and Chauhan, Jagmohan (2025) EchoMotion: enhancing exercise analysis with acoustic sensing. European Signal Processing Conference, Isola delle Femmine, Palermo, Italy. 08 Sep - 12 Oct 2025. pp. 1772-1776 .

Abstract

EchoMotion revolutionizes exercise monitoring with its innovative acoustic-based smart speaker system, designed to perform human pose estimation using inaudible acoustic signals. Leveraging a combination of acoustic features, deep learning techniques, and a custom loss function, the system transforms acoustic reflections from the human body into precise 3D pose estimations. Ground truth data were recorded using the Microsoft Azure Kinect DK, a depth-sensing camera used for evaluation. Data were collected from 22 participants performing five fast-movement cardio exercises in both home and lab environments, yielding over 11 hours of synchronized acoustic and ground truth data. EchoMotion achieved a low Mean Absolute Error (MAE) of 0.59 mm, demonstrating superior accuracy for fast movement exercises compared to the reported MAE range of 2.8 mm to 96 mm in SOTA works, which also focus on slow movements. Our system is non-invasive, cost-effective, respects privacy, and is capable of performing in various acoustic conditions, making it an ideal tool for home-based exercise monitoring and feedback. EchoMotion’s ability to analyze the exercise pose estimations provides valuable insights for users, trainers, and clinicians, enhancing the quality of remote exercise programs.

This record has no associated files available for download.

Contributors

Author: Mohammed Mosuily

Download statistics