| Summary: | Articulation and phonological disorders are associated with the problems of speech sound production and the improper usage of phonological rule in a language respectively. In a conventional assessment, client with articulation and/or phonological disorder is diagnosed manually by speech-language pathologist. Some problems of conventional assessment are identified such as subjectivity in transcription of speech sounds, no integration between the assessment tests of articulation evaluation, phonological evaluation, stimulability test and auditory discrimination test, expensive and non user-friendly instrumental devices and lack of computer-based Malay assessment. Thus, the study proposes an effective methodology for computer-based Malay articulation-phonological assessment in order to overcome the problems. The study also proposes three modular neural networks to recognize the Malay speech sounds in an objective way based on speech characteristics: place of articulation, manner of articulation and voicing. The training and testing of the neural networks involve 60 and 32 Malay children with a database size of 15,668 speech samples. The computer-based assessment requires a client to speak to a microphone, without any attached physical devices. All the results of speech sound transcription and analysis are summarized by the computer-based system. The proposed computer-based assessment system examines 22 Malay consonantal phonemes in isolated words, at three word positions: initial, middle and final. Clinical trial involving 14 Malay primary school children and three children with hearing and speech impairment reveals that the proposed computer-based assessment is valid and reliable. The computer-based system is able to achieve an accuracy of 65% in transcribing the speech sounds in a speaker-independent manner. Though the recognition rate is not very promising to make the system a total computer-based assessment, the proposed system is a helpful analysis tool for clinician, which is able to provide the results in a click of time.
|
|---|
