Dysarthric Speech Reconstruction

Dysarthric speech reconstruction (DSR) aims to convert unintelligible or distorted speech caused by neurological disorders into clear, natural-sounding speech. Current research heavily utilizes multi-modal approaches, incorporating visual information (lip movements) alongside audio, and employs neural network architectures like neural codec language models and encoder-decoder frameworks, often incorporating techniques like adversarial training and self-supervised learning to improve speaker similarity and prosody. These advancements significantly improve speech intelligibility and naturalness, offering substantial potential for enhancing communication and quality of life for individuals with dysarthria.