DSSCNet: A Transfer Learning Framework for Cross-Corpus Dysarthric Speech Severity Classification

DSSCNet is a newly introduced deep learning framework designed for cross-corpus dysarthric speech severity classification, utilizing transfer learning and multi-corpus learning techniques. The model was pre-trained on one dysarthric speech corpus and fine-tuned on another, achieving 75.80% accuracy on the TORGO dataset and 68.25% on UA-Speech, significantly outperforming existing models. This advancement is crucial for practitioners as it enhances the robustness of automated assessments for dysarthria, addressing challenges like speaker variability and class imbalance in speech data.