Text-to-Speech Bahasa Melayu Dialek Lubuk Kandis Menggunakan Metode VITS

Abstract

The existence of numerous regional languages in Indonesia reflects the country's cultural richness, with their historical distribution forming various dialects. One of them is the Malay Language of the Lubuk Kandis Dialect. To support the preservation and dissemination of this language, this study applies speech synthesis technology in the form of Text-to-Speech (TTS) to generate speech output for the Lubuk Kandis dialect. A dataset consisting of 500 daily activity sentences was recorded by a native speaker of Lubuk Kandis and validated by a local linguistic expert. The Variational Inference with Adversarial Learning for End-to-End Text-to-Speech (VITS) method was adopted to produce speech output that resembles natural human voice. The evaluation used the Mean Opinion Score (MOS) test involving five respondents and resulted in an average score of 4.72, indicating that the generated speech falls into the good category. These findings demonstrate that the VITS method is effective in synthesizing natural and intelligible speech for the Malay Language of the Lubuk Kandis Dialect, providing potential support for the preservation of regional languages.