An Improved Syllabification for a Better Malay Language Text-to-Speech Synthesis (TTS)
Abstract
Text-to-speech (TTS) is an important component of robots, humanoids and Internet of Things as a mean of human computer interaction. One of the main components of TTS is text processing that functions as the producer of syllabic speech units to be used in the generation of human-like speech. The naturalness of TTS largely depends on text processing component, particularly word syllabification. Syllabification is a process of segmenting the given text input into sequence syllabic speech units. This paper begins with investigations of previous syllabification technique of Malay language to identify the limitations. An improved syllabification technique is then proposed and compared against the performance of another three known syllabifications. The datasets used comprises 25,000 words collected from Malay language online national newspaper articles and Wikitionary Open Content Dictionary. Word Error Rate (WER) percentage is calculated and our proposed syllabification technique achieved the lowest WER of 2.61% with an accuracy rate of 97.39%.