Phoneme Based Respiratory Analysis of Read Speech

Recent work shows that it is possible to use deep learning techniques to sense the speaker's respiratory parameters directly from a speech signal. This can be a beneficial option for future telehealth services. In this paper, we dive deeper and study how respiratory effort depends on the linguistic content of the speech utterance. This is obtained by analysis of respiratory belt sensor data and phoneme-aligned speech data. The results show, for example, that the respiratory effort was highest for fricatives, compared to other broad phonetic classes, and especially high for the glottal consonants. The insights may help to develop more efficient protocols for respiratory health monitoring in telehealth applications.