Study and Implementation of Prosody Manipulation Method For Indonesian Speech Synthesis System

摘要

Speech Synthesis System is a system used to convert text in a language into a sound. The focus of this research is to produce a “humanization” of speech synthesis system pronunciation. The main requirement for Text To Speech system in this research are eSpeak, MBROLA idl database for Indonesia, Human Speech Corpus database which derived from the website that summarizes the words with the most common words used in a country, and three basic types of emotion or intonation that are designed for happy emotions, angry emotions, and sad emotions. The approach method used to develop an emotional filter is to manipulate prosody values (especially pitch and duration values) using predetermined level factors. The test results of Human Speech Corpus perception test for happy emotions are 95%, angry emotion is 96.25% and emotion sad is 98.75%. For the aspect of the clarity test, the audible sound accuracy with the original sentence is 93.3%, and for the clarity level each sentence is 62.8%. For the naturalness aspect to test the accuracy of emotional selection is 75.6% with every happy emotion is 90%, angry emotion is 73.3% and sad emotions of 60%.