PENGENALAN SUARA UCAPAN SUKU KATA BAHASA LISAN BAHASA INDONESIA DENGAN MENGGUNAKAN CIRI LPC, MFCC DAN JARINGAN SYARAF TIRUAN
| Main Authors: | , Abriyono, S.Kom, , Drs. Agus Harjoko, M.Sc., Ph.D. |
|---|---|
| Format: | Thesis NonPeerReviewed |
| Terbitan: |
[Yogyakarta] : Universitas Gadjah Mada
, 2012
|
| Subjects: | |
| Online Access: |
https://repository.ugm.ac.id/97863/ http://etd.ugm.ac.id/index.php?mod=penelitian_detail&sub=PenelitianDetail&act=view&typ=html&buku_id=54521 |
Daftar Isi:
- Voice is one of effective and convinienced communication�s medium among human. However, the used of voice is not only for communication among human but also has another role nowadays. Voice becomes communication medium for human and computer (machine). This fact can be observed through many applications likes robot, home appliances, voice-text conversion software in specific language, and many more. In this research, researcher focus on the recognition of Indonesian word. The research for Indonesian word recognition was still at low amount. The word used for research was at small amount too. In this research, researcher divides the word into the speech syllable. The aim for the dividing system is to reduce the large amount of the word, but still cover all of the word. Researcher found and used 1741 speech syllables in this research. Researcher used several approaches in managing the recognition. The approaches are 11025 Hz, Mono, 8 bit for recording, pre-emphasized, segmentation, framing, and windowing for preprocessing, LPC and MFCC for the features, and back-propagation neural network for the identifier. Used training data in this research was recorded voice from three person which was pronounced twice every person for each of the Indonesian�s speech syllables. For the testing data, researcher used different recorded voice from training data but was pronounced by the same person. The testing data recorded one pronounciation every person for each of the speech syllables. The result using this approach was not reached good performance. The best result performed 0.65% by using MFCC feature.