Towards Heart Rate Categorisation from Speech in Outdoor Running Conditions

• Published in: 2022 E-Health and Bioengineering Conference (EHB) pp. 1 - 5
• Main Authors: Gebhard, Alexander, Amiriparian, Shahin, Triantafyllopoulos, Andreas, Kathan, Alexander, Gerczuk, Maurice, Ottl, Sandra, Dieter, Valerie, Jaumann, Mirko, Hildner, David, Schneeweiss, Patrick, Rosel, Inka, Krauss, Inga, Schuller, Bjorn W.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 17-11-2022
• Subjects: Feature extraction; Heart rate; heart rate classification; Machine learning; Measurement; outdoor running; Pipelines; Support vector machines; Training
• ISSN: 2575-5145
• DOI: 10.1109/EHB55594.2022.9991421

The heart rate (HR) provides key information about the intensity of the cardiorespiratory workout, the level of exertion, and the overall heart condition. In sports, and especially running, tracking the HR and other metrics to monitor training progress and avoid injuries has been recently gaining momentum - a trend titled smart exercising. However, especially for beginners, it can be difficult to properly interpret a metric such as HR, which is why an expert categorisation can be beneficial. Furthermore, it can be uncomfortable to put on multiple wearable sensors or buy extra gadgets for measuring the HR during a running session. In order to tackle these issues, we propose a machine learning pipeline for the prediction of various HR categories based solely on speech samples recorded by a smartphone in outdoor running conditions. To this end, we first extract data representations utilising fine-tuned Transformers, pre-trained convolutional neural networks, and conventional, interpretable feature extraction methods. Afterwards, we apply synthetic feature augmentation on all feature sets to cope with potential class imbalance problems. Finally, we train and optimise various linear support vector machine (SVM) and feed forward neural network (FFNN) models on the obtained and augmented features. The results demonstrate the suitability of the proposed machine learning pipeline for automatic speech-based HR classification.