Towards Voice and Query Data-based Non-invasive Screening for Laryngeal Disorders

Topic of the research is exploration and fusion of non-invasive measurements for an accurate detection of pathological larynx. Measurements for human subject encompass results of a specific survey and information extracted by openSMILE toolkit from several audio recordings of sustained phonation (vowel /a/). Clinical diagnosis, assigned by medical specialist, is a target attribute for binary classification into healthy and pathological cases. Random forest (RF) is used here as a base-learner and also as a meta-learner for decision-level fusion. Fusion combines decisions from ensemble of 5 RF classifiers built on 3 variants of audio recording data (raw and after two types of voice activity detection) and 2 variants of questionnaire (with 9 and 26 questions) data. Out-of-bag equal error rate (EER) was found to be higher for audio data and lower for query, but each variant was outperformed by the fusion where the lowest EER of 4.8% was achieved. Finally, due to noteworthy performance of the query data, 22 association rules (11 healthy + 11 pathological) using 17 questions were obtained for comprehensible insights. Key–Words: ensemble methods, random forest, affinity analysis, voice activity detection, voice pathology detection