2020/04/04 by Lam Pham, Huy P. Phan, Pham, Lam +7
Computer Science · Medicine · #Audio and Speech Processing (eess.AS) #FOS: Computer and information sciences #FOS: Electrical engineering #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Music and Audio Processing #Phonocardiography and Auscultation Techniques #Respiratory and Cough-Related Research #Sound (cs.SD) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2004.04072
openalex publication_date 2020/04/04 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
This paper presents and explores a robust deep learning framework for\nauscultation analysis. This aims to classify anomalies in respiratory cycles\nand detect disease, from respiratory sound recordings. The framework begins\nwith front-end feature extraction that transforms input sound into a\nspectrogram representation. Then, a back-end deep learning network is used to\nclassify the spectrogram features into categories of respiratory anomaly cycles\nor diseases. Experiments, conducted over the ICBHI benchmark dataset of\nrespiratory sounds, confirm three main contributions towards respiratory-sound\nanalysis. Firstly, we carry out an extensive exploration of the effect of\nspectrogram type, spectral-time resolution, overlapped/non-overlapped windows,\nand data augmentation on final prediction accuracy. This leads us to propose a\nnovel deep learning system, built on the proposed framework, which outperforms\ncurrent state-of-the-art methods. Finally, we apply a Teacher-Student scheme to\nachieve a trade-off between model performance and model complexity which\nadditionally helps to increase the potential of the proposed framework for\nbuilding real-time applications.\n