Efficacy of Current Dysarthric Speech Recognition Techniques

• Published in: 2023 International Conference on Advanced Computing & Communication Technologies (ICACCTech) pp. 657 - 663
• Main Authors: Malik, Medha, Khanam, Ruqaiya
• Format: Conference Proceeding
• Language: English
• Published: IEEE 23.12.2023
• Subjects: Automatic Speech Recognition (ASR); Data models; Dysarthric Speech Recognition (DSR); Gaussian Mixture Models (GMMs); Hidden Markov models; Hidden Markov Models (HMMs); Linear Predictive Coding (LPC); Mel frequency cepstral coefficient; Mel Frequency Cepstral Coefficients(MFCC); Perceptual Linear Prediction (PLP); Predictive coding; Predictive models; Speech enhancement; Speech recognition
• DOI: 10.1109/ICACCTech61146.2023.00111

Speech Recognition system development is critical to supporting good communication. Furthermore, since speech and language technology progress on a daily basis, advancements in Dysarthric Speech Recognition systems are advantageous to those with dysarthria. In this work, we look in depth at the primary techniques employed in Dysarthric Speech Recognition. It starts with a quick description of dysarthria and how it affects speech patterns. It then analyses several ways to character extraction. Some of these methods are Mel Frequency Cepstral Coefficients, Linear Predictive Coding, Perceptual Linear Prediction, and Gammatone Frequency Cepstral Coefficients. Various ways for determining differentiating aspects of dysarthric speech have also been developed. These include techniques such as formant re-synthesis and acoustic space manipulation. The goal of these strategies is to improve speech recognition and intelligibility by enhancing various acoustic elements of speech. This study examines current advancements and problems in Dysarthric Speech Recognition in depth. Deep learning approaches and improvements to speaker and lexical models are being researched as strategies to increase DSR system efficiency and accuracy. Despite advancements, there are still a number of hurdles in the field of dysarthric speech recognition. Gathering data, coping with noise interference, building specialist models, and reacting to variances in speech are all examples of challenges faced in DSR model. This research also considers potential future directions for DSR, such as the utilisation of assistive technology and multimodal approaches. It also discusses advanced methodologies for feature extraction and user accessibility.