The Indonesian Language speech synthesizer based on the Hidden Markov Model

Speech synthesizer is a technology which gives the computer a capability to speech text sequences. In this research, we develop a speech synthesizer for Indonesian Language based on the Hidden Markov Model (HMM). The Speech synthesizer using the HMM can produce more appropriate result than based on the syllable concatenation. There are some studies of speech synthesizers using HMM for Indonesian Language. However, it still has some problems such as it still cannot distinguish between vowel "e" ("e" in "get" is different from "e" in "apple"); It cannot handle abbreviation, numbers, special characters, and foreign (English) terms widely. In this research, we also proposed some methods to solve those problems. To solve "e" problem, this research divided the HMM for the 2 "e" vowel. To solve the other problems, the "e" rules, the abbreviation rules, the number rules, the special character rules, and the foreign term rules are made. To evaluate the synthesizer, we employ two methods: the Mean Opinion Score (MOS) to measure the naturalness of synthesized speech; and the Semantically Unpredictable Sentence (SUS) to measure the accuracy of the synthesized speech. Result shows that the developed speech synthesizer improved the naturalness of synthesized speech. It achieves 4.1 for MOS point and 96,07 % word accuracy.

I. Introduction

Nowadays, we interact with computer using keyboard, mouse, and touch screen. However, typing for computer input is not natural for human interaction. We also usually see the computer screen if the output is visual. Some research focused on how to increase the naturalness of interaction. Speech as our most of natural means communication can be used when interacting with computer. Some of the technologies employing speech such as voice recognizer and automatic speech synthesizer have been developed for many languages, especially English [1].