“Multi-Modal Dataset Across Exertion Levels: Capturing Post-Exercise Speech, Breathing, and Phonocardiogram” was accepted and presented at SenSys’25

We are excited to share that Jingping presented our work, “Multi-Modal Dataset Across Exertion Levels: Capturing Post-Exercise Speech, Breathing, and Phonocardiogram,” at SenSys’25 during CPS-IoT Week 2025. This dataset and study were developed through a collaborative effort by lab members Jingping, Yuang, Minghui, Ziyi, and Runxi.

In this work, we introduced a first-of-its-kind multi-modal audio dataset capturing physiological changes in speech, breathing, and heart sounds after cardio exercise. Existing datasets largely focus on resting conditions and fail to reflect the dynamic variations that occur post-exertion, such as speech disfluencies, altered breathing rhythms, and varying heart sound intensities. To address this gap, we recruited 59 diverse participants and collected 250 sessions of post-exercise data, including structured reading, spontaneous speech, breathing sounds, and phonocardiogram (PCG) signals. We also conducted a case study using the speech audio in the dataset for exertion level classification. This dataset provides a valuable foundation for developing robust models in speech and cardiorespiratory monitoring that perform reliably under physically demanding conditions.

The dataset is publicly available and can be accessed here:
🔗 https://github.com/Columbia-ICSL/data_after_cardio/tree/main