Deep Learning-based Dysarthric Speech Detection

Description

• Learning rate is halved after 5 epochs without improvement in validation loss.
• Trained on frames and Tested on Audio files using soft voting.

Reference

• Experimental investigation on STFT phase representations for deep learning-based dysarthric speech detection
• Temporal Envelope and Fine Structure Cues for Dysarthric Speech Detection Using CNNs

Accuracies

Dataset	Description	Accuracy
Torgo	Dysarthric 34.89% vs. Control 65.11%
UASPEECH