KAJIAN TENTANG PENGGUNAAN CEPSTRAL UNTUK MENENTUKAN INTERVAL FONEM DALAM UCAPAN BAIIASA INDONESIA SECARA OTOMATIS

• Main Author: Parminto
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/5134
• Institution: Institut Teknologi Bandung

<b>Abstract:<p align="justify"> <br /> Speech data base development is considered as one of the most time-consuming activity within the development of a Text-To-Speech (TTS) system that uses concatenation technique. The main process employed by the activity is the determination of phoneme intervals in speech signal recording, which is used as a sample.<p align="justify"> <br /> The objective of this research is to study the possibilities of using cepstral coefficient to automatically determine phoneme interval in Indonesian language speech. The main principle that is used in this research is to find out the smallest difference between the (known) reference phoneme's cepstral and the cepstral of phonemes available in speech signals. The cepstral difference is calculated by moving the reference phoneme's cepstral along the cepstral of word speech being tested. The phoneme location and interval then determined based on location that has the smallest cepstral difference value.<p align="justify"> <br /> In this research, a system was developed using the Matlab version 5.3 software to carry out the work. The phonemes being tested are only limited to the Indonesian language vowels, which are a, i, e and o.<p align="justify"> <br /> Result of the test on 200 words, which includes 536 phonemes, proved that cepstral coefficient can be used to determine phoneme intervals in Indonesian language speech. The system successfully identifies all phonemes in 158 words (79%). There are 490 phonemes (91.42%) out of a total of 536 phonemes within the 200 words that were tested, were successfully identified. Failures are generally caused by the existence of similarities between one phoneme and another.