VoiceTransformer pour la détection précoce des maladies neurodégénératives
Résumé
This work explores the use of attention-based models (Transformers) for automatic feature extraction from vocal data to detect neurodegenerative diseases. These models capture long-term dependencies in speech sequences, making them suitable for identifying early signs of these pathologies. By fine-tuning pre-trained encoders on patient data, the study assesses performance compared to traditional acoustic descriptors and machine learning models. The findings highlight the potential of attention-based approaches for enhancing non-invasive diagnostic systems using voice analysis.