Aplikasi Pengenalan Ucapan dengan Ekstraksi Mel-Frequency Cepstrum Coefficients (MFCC) Melalui Jaringan Syaraf Tiruan (JST) Learning Vector Quantization (LVQ) untuk Mengoperasikan Kursor Komputer

*Angga Setiawan  -  JURUSAN TEKNIK ELEKTRO, Indonesia
Achmad Hidayatno  -  JURUSAN TEKNIK ELEKTRO, Indonesia
R. Rizal Isnanto  -  JURUSAN TEKNIK ELEKTRO, Indonesia
Diterbitkan: .
Akses Terbuka

Citation Format:
Article Info
Bagian: Artikel - Teknik Tenaga Listrik
Bahasa: EN
Statistik: 1126 2561
Sari

During this time, computer cursor operation was done by pressing and moving the mouse. So, this is less flexible for computer user that require movement in operating a computer, since to use mouse comfortably someone has to sit. Moreover, physical completeness is required for mouse operating, so that for someone who has physical disabilities feels difficult to operate it. Therefore, it is required to develop a system that provides a better comfort and flexibility not only for the healthy user computer but also for the user computer who has physical disabilities. In this final project, computer cursor operation program via voice is created. With this program, someone will have more flexibility when operating the computer cursor and also people with physical disabilities is enabled to communicate with computer. Voice recognition is a technology that is apllied in this program, with the feature extraction process used MFCC (Mel-Frequency Cepstrum Coefficients) method. As for the recognitions process used artificial neural network type LVQ (Learning Vector Quantization). Voice is passed through a microphone and then it is analyzed by MFCC to produce MFCC coefficients. These coefficients are used as input vector for LVQ neural network and used as data to train the network until it has the classification capability. Programming language that is used in creating this software is Delphi programming language. Based on the result of the testing program, it is found that the success percentage rate of voice recognition with training data, that is data which is derived from databases that have been recorded and trained into the program which amounts to 240 data, is 88,89 %. While in the testing with test data, that is data which is derived from the real time sayings of respondents which is amounts to 240 data, it is found that the success percentage rate of voice recognition is 83,99 %.

Keyword : voice recognition, computer cursor, MFCC, LVQ

Article Metrics: