Development of Malay Word Pronunciation Application using Vowel Recognition
International Journal of u- and e- Service, Science and Technology
In Malaysia, many researchers focus on developing speaker independent systems for training or articulation therapy or to assist language learners to learn about Malay Language or Bahasa Malaysia. Accuracy, noise robustness and processing time are concerns when developing speech therapy systems. In this study, a Malay word pronunciation test application was developed using the first 3 format and fundamental frequencies in an effort to improve pronunciation in Malay. This application was
... cation was developed using Matlab and uses a vowel recognition algorithm classified using MLP classification technique. The application was developed and tested on UUM undergraduate students. For vowel classification, when fundamental frequency was added, 3-format feature vowel classification rate increased by 1.55% for male gender and 1.48% for female. When combined both genders, a more significant improvement of 1.71% was seen. The developed pronunciation application test results showed that the pronunciation application can assist in testing and improving their Malay word pronunciation. It was also observed that, vowel /i/, /e/, /o/ and /u/ are often mispronounced due to pronunciation habits. In Malaysia, Universiti Kebangsaan Malaysia (UKM) has two computer-based speech therapy systems situated in the Clinic of Audiology and Speech Sciences. They are the Kay Elemtrics VisiPitch and IBM Speech Viewer  . These systems are used for voice therapy, but not used for training or articulation therapy. Furthermore, these systems use English speech therapy. There are other applications like OLTK (Optical Logo-Therapy Kit)  and VATA (Vowel Articulation Training Aid)  . These systems have limitations, and not robust enough to handle real-time identification of vowels. In 2007, Tan et. al  developed a Malay Speech Therapy Assistance Tool (MSTAT) which is used to assists therapists in diagnosing children for language disorder and train the children suffering from stuttering problem. It uses speech technologies consisting of speech recognition, Malay Talking Head and Malay text-to-speech system. A Computer-based Malay Language Articulation Diagnostic System was developed using Hidden Markov Model (HMM) and Mel-Frequency Cepstral Coefficients (MFCCs)  . It was developed using a database of Malay words. In 2012, Tan et.al developed a Malay dialect translation and synthesis system, but still at a preliminary stage  . The speech synthesis system used here is an HMM speech synthesis system (HTS Speech Synthesis System) at a sampling rate of 22 kHz. The results were promising, but the system does not test on pronunciation. A research was done in 2014 with the objective of developing an ASR system for Malay speaking children  . The speech corpus comprises of six children uttering a total of 390 sentences. The parameter training is performed using the HTK toolkit by utilizing an HMM speech acoustic model of Malay speaking children. The system can accurately recognize of up to 76% of test words. Yusof et.al did a study about speech intelligibility of deaf children in Malaysia using a Malay Speech Intelligibility Test (MSIT) system  . Researchers from Universiti Malaysia Sarawak did a study on syllabification algorithm based on Malay syllable structure  . It was used to build the Iban and Bidayuh syllable list and speech corpus. The accuracy, using Categorical Estimation (CE), gave a mean score of 3.07 out of 5.