Intelligibility assessment of impaired speech using Regularized self-representation based compact supervectors. (July 2022)

• Record Type: Journal Article
• Authors: Malini, S.; Chandrakala, S.
• Is Part Of: Computer speech & language. Volume 74(2022)
• Publication Date: 2022-07
• Subjects: Intelligibility assessment -- Impaired speech -- Neurological disorders -- Universal Background Model (UBM) -- Adapted Gaussian Mixture Models (AGMMs) -- Supervectors -- Regularized self-representation (RSR) -- Disordered speech recognition -- Support vector machines (SVM) -- Neurological disorder -- Speech assisted system -- Mel-frequency cepstral coefficients; Speech processing systems -- Periodicals; Automatic speech recognition -- Periodicals; Computers -- Periodicals; Linguistics -- Periodicals; Speech-Language Pathology -- Periodicals; Traitement automatique de la parole -- Périodiques; Reconnaissance automatique de la parole -- Périodiques; Automatic speech recognition; Speech processing systems; Electronic journals; Periodicals; 006.454
• DOI: 10.1016/j.csl.2022.101355
• Languages: English
• ISSNs: 0885-2308

Abstract

Abstract: Automatic intelligibility assessment of impaired speech is essential and helps in assisting people with speech impairment in therapy and clinical diagnosis. It is important to learn compact and discriminative representations to assess the quality of impaired speech. We propose a regularized self-representation (RSR) based compact supervectors for intelligibility assessment of the impaired speech. Supervectors of huge dimension are used as base representations to obtain a regularized self-representation based compact supervectors. 10-digits and 100-common words datasets of the UA-SPEECH database are used to evaluate the performance of the proposed RSR-based compact supervector approach. The above two datasets have four intelligibility levels: high, medium, low, and very low. The proposed RSR-based compact supervector approach performs significantly better than the hybrid GMM/SVM, supervector, x-vector, i-vector, and Bag of Models-based approaches for both datasets. Compactness and discriminative ability of the proposed representation help to improve the performance of intelligibility assessment of the impaired speech.