Dual Cross-Attention Parkinson’s Disease Classification Using Vocal Feature Sets

Parkinson’s disease is a disabling condition that affects the quality of life of individuals with both motor and non-motor symptoms. Auditory disorders are one of the non-motor symptoms that individuals will face in the early stages of the disease. Therefore, the use of vocal features for early detection of Parkinson’s disease is a crucial aspect of diagnosis. This paper presents a novel approach to Parkinsonts disease diagnosis using vocal features and advanced neural network architectures. In this study, a structure based on dual cross-attention is introduced that combines features extracted from time-frequency representations, specifically the Wavelet Transform (WT) and Tunable Q factor Wavelet Transforms (TQWT), describing both temporal and frequency features of the audio signal simultaneously. Ultimately, a self-attention block is responsible for determining the class assignment. The results demonstrate that the proposed network performs well compared to existing networks in the context of Parkinson’s disease diagnosis.