Heart rate extraction from speech signals

The heart rate signal is a crucial physiological marker often used to identify multiple conditions and diseases such as chronic stress, depression, heart failure, and arrhythmias. Apart from its high diagnostic power, the heart rate signal is relatively easy to record using smart watches or electrode-based systems that measure the electrocardiogram (ECG). However, one still relies on the good use of such systems. For instance, the electrodes must be correctly attached or the watch needs to be completely fixed to the wrist so that movement artefacts do not influence the recordings. As a consequence, these systems are often limited to either subjects who willing use them (e.g. as gadgets) or who have sufficient motoric and cognitive skills to use them regularly. With this in mind, the extraction of heart rate information from signals like speech is of paramount importance. Reasons for this include the fact that speech is an easier to record signal, which can be acquired under (often) general conditions.

Assignment

In this project, students will first record speech and ECG signals simultaneously using available hardware. The speech signals will need to be processed and different algorithms will need to be designed and tested in order to extract heart rate information from them. This information will be validated against the heart rate signal extracted from the ECG signals. Different conditions can be tested, for instance, when subjects only produce certain sounds, or when they are in different rooms (e.g., dead room vs. any classroom).

Requirements

• Matlab
• Courses on Signals & Systems and on Machine learning

Contact

dr. Carolina Varon

Signal Processing Systems Group

Department of Microelectronics

Last modified: 2020-03-13