Overview
The voice pathology detection project aims to create a method for the simple, non-intrusive, and reliable detection of pathologies in the human vocal cords. Different voice pathologies caused by lesions on the vocal cords affect up to 9% of the population. While patients usually notice minor changes in voice parameters, such as hoarseness, these pathologies can originate from a wide spectrum of causes, from the common cold to a malignant tumor.
Diagnosis of voice pathologies currently requires invasive endoscopy procedures, such as laryngostroboscopy or surgical microlaryngoscopy. Expert otolaryngologists can detect different pathologies from patients' speech. Unfortunately, the current classification rate for human experts is only about 60-70%.
Our project's objective is to aid these procedures with computer-based diagnostic tools. The tools will use non-intrusive data such as speech recordings and questionnaires. Different usage scenarios are considered, from supporting clinic-based professional otolaryngologists to large-scale screenings of high-risk subjects at their workplaces to increase the probability of early detection.
The project is based on collaborations with two Lithuanian research teams: researchers from the Ear, Nose, and Throat Clinic at the Lithuania University of Health Sciences (LUHS), led by Prof. Virgilijus Ulozas, and the machine learning group at the Kaunas Technological University (KTU), led by Prof. Antanas Verikas.
One of our key research tasks focuses on processing the speech data and extracting meaningful features useful for classification. For this task, our team leverages its deep and well-established experience in speech processing, speech analysis, and speech classification.