On the Use of Bronchial Breath Sounds for Person Identification

• Published in: Journal of Information Science and Engineering Vol. 37; no. 1; pp. 219 - 241
• Main Authors: VAN-THUAN TRAN, 林勇志(YUNG-CHIH LIN), 蔡偉和(WEI-HO TSAI)
• Format: Journal Article
• Language: English
• Published: Taipei 社團法人中華民國計算語言學學會 01.01.2021; Institute of Information Science, Academia Sinica
• Subjects: Accuracy; Acoustics; Bronchi; Glottis; Sound; Sound recording; Stochastic models; Support vector machines; Trachea; Gaussian mixture model (GMM); i-vector; bronchial breath sound; person identification; MFCCs
• ISSN: 1016-2364
• DOI: 10.6688/JISE.202101_37(1).0014

Bronchial breath sound is the sound of turmoil flow produced by the inspiratory air through glottis, trachea or major bronchi. It cannot be only used to diagnose the respiratory tract and lung-related diseases but also used to distinguish one person from the other and thereby identifying patients since it contains personal physiological characteristics. This study captures the bronchial breath sound by using a stethoscope attached on a subject's neck. For each person, the Mel Frequency Cepstral Coefficients (MFCCs) are computed for his/her bronchial breath sounds, and then represented by a stochastic model. Given an unknown breath sound recording, the proposed person identification system determines who among a set of candidate people produced the breath sound by matching the MFCCs of the sound to each of the stochastic models. Furthermore, we apply the i-vector approach in the system to boost the identification accuracy. To evaluate the generality of our experimental results, we additionally utilize other general identification schemes including support vector machine, random forest, and naive Bayes. Our experiments conducted on a dataset composed of 8 persons show that the accuracy of identifying people from their breath sounds can attain 92%.