Design and development of voice transformation algorithms for improved speech in aphasic individuals

Abstract

Aphasia is a neurological disorder caused due to damage in different regions of human brain, which are mainly responsible to perform different executive functions, like language, memory, and attention. In the United States (US), it has been determined that since 2020, about 1,80,000 people suffered from aphasia. In general, aphasia patients suffer from language-related issues as well as speaking, listening, reading, and writing skills. The verbal production of persons suffering from aphasia is more complex and tricky to understand because of language issues, like the construction of incorrect, omitted words or paraphasia sentences. Aphasia causes severe impacts on the life of affected persons, where it also comes under motor control disorders, such as dysarthria and apraxia. The aphasia results in language disorders together with atypical prosody and articulatory distortions. The impacts of aphasia also cause different components and levels including semantics, syntax, lexicon, and phonology. Generally, the speech of persons is analyzed to determine the severity or type of aphasia. The narrative spontaneous speeches generated by a person s aphasia are evaluated to detect the severeness of disorder. The subjective listening is used to analyse the speech intelligibility for the estimation of word percentage accurately that are understood by people. However, these processes are expensive and labor intensive, which is also caused by the perception of listener s. In recent years, automatic speech recognition systems have been utilized to transcribe speech by extracting textures. Deep learning techniques are utilized to automatically increase speech quality by utilizing unknown and known noise sources. In the first contribution, a technique for speech intelligibility for aphasia is devised using the Gradient Tangent Search Optimization (GTSO) algorithm-enabled voice transformer newline