Automated Speech Correction Assistive Technology for Malayalam Articulation Errors

Individuals with speech impairments confront substantial challenges affecting their communication capabilities and overall quality of life. Among these, articulation errors pose a significant barrier to effective communication, often resulting in frustration and social isolation. The articulation process entails the modification of sound waves generated by the airstream, phonation, and oral-nasal processes [Glossary of Linguistic Terms]. Speech production occurs in the oral cavity, involving key organs such as the tongue, upper and lower teeth, upper and lower lips, and elements of the roof of the oral cavity (e.g., alveolar ridge, palate, and velum). These organs can act as either active or passive articulators. Speech sounds are differentiated based on their place and manner of articulation. Articulation errors arise from disorders affecting these specific attributes. The present paper introduces an assistive technology to enhance Malayalam speech impacted by articulation errors. Malayalam is primarily spoken in southern India and has its phonological complexities. The technology employs a rule-based system for real-time analysis of the speaker's speech to offer targeted assistance. The system has been trained on a dataset of Malayalam speech samples with articulation errors collected from an individual diagnosed with cerebral palsy. The interface is user-friendly and web-accessible. Experimental evaluations corroborate the system's efficacy, evidenced by a Word Error Rate (WER) of 8.2% and a Signal-to-Distortion Ratio (SDR) of 20.6. The proposed solution aims to improve the speech capabilities of individuals with articulation errors, bolstering their communication skills and self-assurance in social interactions.