Development of Malay Speech Translation System using Python

Abstract

Translator is a useful tool which allows a person to communicate with other people using different languages that are not familiar with when visiting a foreign place. The existing voice translators experiencing slower process in translating spoken word or sentences from one language to another. The paper presents the development of translation system for translating spoken Malay word or phrases into Chinese and English or vice versa using Python. Mel Frequency Cepstral Coefficient (MFCC) and Hidden Markov Model (HMM) algorithms are proposed in the system to improve the system performance while converting speech to text or text to speech. The reliability of translation and speed performance of the system have been conducted with 15 people, 8 males and 7 females. The reliability test for sample rate 8000 bps, 11025 bps, 32000 bps and 48000 bps indicating the corresponding translation quality are 53%, 73%, 100% and 100% respectively. This suggested that the sample rate 32000 bps and 48000 bps is more reliable and the translation quality is better. The translation speed performance for sample rate 8000 bps, 11025 bps, 32000 bps and 48000 bps are 5.73s, 5.81s, 5.87s and 5.97s respectively. Although sample rate 8000 bps has resulted fastest translation speed, but its corresponding system quality is poor. As conclusion, a Malay speech translation system is successfully developed and tested. The system can accurately translate the Malay words in relatively short time. Thus, the result suggests that the most suitable system operational sample rate is 32000 bps, because it result in faster and reliable translation.