Dysarthric Speech Recognition

Dysarthric speech recognition (DSR) aims to develop automatic speech recognition systems capable of accurately transcribing speech from individuals with dysarthria, a motor speech disorder. Current research focuses on mitigating the challenges posed by high inter-speaker variability and limited training data, employing techniques like prototype-based adaptation, speaker-adaptive fine-tuning of large language models (e.g., Whisper), data augmentation (including text-to-dysarthric-speech synthesis and GAN-based methods), and the incorporation of speaker-specific features like impairment severity. Advances in DSR hold significant potential for improving communication accessibility and quality of life for individuals with dysarthria, and are driving innovation in areas such as self-supervised learning and cross-lingual adaptation of speech models.