A Preliminary Study on Vowel Recognition via CNN for Disorder People in Malay Language

Abstract

Stroke is one disease showing an increment trend as people live their lives in a stressful manner. Rehabilitation is one of the procedures to recover the patient to a normal condition. The rehabilitation process and activities require an extended period to retrain back the patient's capability, speak, listen, walk, etc. For this, a dedicated physiotherapy procedure was conducted according to the rehab trainer and expertise. One of the rehabilitations is to help the patient to have back their speaking skill and capability. The rehabilitation activities are generally conducted manually through manual listening and teaching the stroke patient periodically by the rehab trainer. The manual rehabilitation activities physically require the rehab trainer's presence, documentation, and manual data recording. This manual activity could be challenging when we face a lack of trainers and the situation of many patients with less trained in the field. Therefore, an intelligent system could be an alternative for rehabilitation to provide the user-friendly and straightforward technique to learn, repeat, and evaluate. In the paper, as the preliminary study, we proposed a smart vowel recognition for Malay Language using Convolutional Neural Network (CNN). We also proposed a new Malay Language dataset consist of 5 vowels, /a/, /e/, /i/, /o/ and /u/ for the use of future research. The result shows that the vowel recognition using this dataset is comparable and suitable for recognizing the vowel type.