Systematic Study of Dysarthric Speech Recognition: Spectral Features and Acoustic Models

This study presents a systematic investigation into dysarthric speech recognition, focusing on the integration of various acoustic features with different Acoustic Models. Notably, the use of Pitch features enhanced recognition performance, leading to a 4.65% improvement in isolated word recognition and a 4.63% improvement in sentence recognition using the Factorized Time Delay Neural Network (F-TDNN) model. These findings highlight the importance of feature selection in addressing the acoustic variability inherent in dysarthric speech, offering valuable insights for practitioners developing speech recognition systems for this population.