Heart Rate from Read-Speech Influenced by Physical Exercise

Heart rate (HR), which is the count for number of times heart beats in a minute and its continuous change over a period of time is called as Heart Rate Variability (HRV), along with the HR itself, provides insights into an individual’s cardiac health. In this paper, we propose Independent Component Analysis (ICA) based technique for detecting HR from speech signals. We have used, two-layered analysis for this purpose. The first layer extracts breathing patterns from the speech signal using a deep pre-built model called - SBreathNet. In the second layer, these speech-derived breathing patterns (SDBPs) are used for the extraction of HR and HRV. We have compared the HR and HRV of the candidates reading a phonetically balanced text in three distinct scenarios: 1) Normal breathing, 2) After deep breathing, and 3) After physical exercise. We observe that the HR for deep breathing activity starts with a lower base-value and that for physical activity starts with a higher base-value than that of the normal breathing. HRV shows higher values in the beginning of reading the passage. Both HR and HRV stabilizes with time. We achieved an average HR error of 5.7 beats-per-minute across seven speakers and three activities.