Hybrid CNN-Bi-LSTM with Temporal Capsule Network for the Detection of Parkinson’s Disease Using Speech Signals
Year: 2026
Type: article
Abstract: Parkinson's Disease (PD) is a progressive neurodegenerative disorder characterized by motor and speech impairments arising from dopaminergic neuronal degeneration.Speech-based analysis provides a non-invasive alternative for automated PD identification, since pathological alterations in phonation, articulation, and vocal stability manifest during the early stages of disease progression.Conventional machine learning (ML) approaches primarily depend on handcrafted acoustic descriptors and shallow classification models, which exhibit limited capability in representing complex spectral-temporal characteristics and hierarchical feature relationships embedded within speech signals.To address these limitations, a hybrid deep learning framework integrating a Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Temporal Capsule Network is proposed for PD detection from speech recordings.Mel-spectrogram representations derived from speech signals serve as the input feature space, where CNN layers extract discriminative spectral patterns, the BiLSTM module models bidirectional temporal dependencies, and the capsule architecture preserves hierarchical relationships through dynamic routing mechanisms.Experimental evaluation was conducted using a publicly available Parkinson's voice dataset comprising 1,134 speech samples.The proposed framework attained an accuracy of 98.24%, macro-average precision of 98.32%, macro-average recall of 98.21%, macro-average F1-score of 98.24%, and a ROC-AUC of 0.9997 on the independent hold-out test set.Ablation analysis and comparative evaluation further demonstrate the contribution of spatial, temporal, and hierarchical feature learning toward classification performance.The experimental results demonstrate that the proposed CNN-BiLSTM-Capsule architecture provides an effective approach for automated speech-based Parkinson's Disease classification on the evaluated public dataset and consistently outperforms the considered baseline models.
Authors: Sreekutty S. D, D. Pamela, Shajin Prince, Jayanthi
Institutions: Karunya University, SRM University
Language: English
Topic: Voice and Speech Disorders
Subfield: Speech and Hearing
Field: Health Professions
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W7213623945