Aplikasi Sistem Neuro-Fuzzy untuk Pengenalan Kata

Authors

  • Yohanes TDS Fakultas Teknologi Industri, Faculty of Industrial Technology, Petra Christian University
  • Thiang Faculty of Industrial Technology, Petra Christian University
  • Suntono Chandra Faculty of Industrial Technology, Petra Christian University

:

https://doi.org/10.9744/jte.2.2.

Keywords:

neural network, neuro-fuzzy, speech recognition.

Abstract

This paper describes implementation of neuro-fuzzy system, a hybrid system between neural network and fuzzy logic. In this research, neuro-fuzzy system is implemented for speech recognition. The words that would be recognized are "nol", "satu", "dua", "tiga", and "empat". The neuro-fuzzy system had one input layer, four hidden layers, and one output layer. The experiment was done by compare neuro-fuzzy system with neural network system. The results showed that neuro-fuzzy system give better result than neural network system. Learning time for neuro-fuzzy system was faster than neural network system. Neuro-fuzzy needed 160,000 iterations for learning 270 sound samples. On the contrary, neural network needed 500,000 iterations for learning 270 sound samples. Neuro-fuzzy system could recognize up to 96,36 %. Neural network system could recognize only 62,86 %.
Abstract in Bahasa Indonesia :

Makalah ini menjelaskan tentang aplikasi sistem hybrid antara neural network dan fuzzy logic yang dinamakan sistem neuro-fuzzy. Dalam penelitian ini, sistem hybrid neuro-fuzzy diaplikasikan untuk pengenalan kata yaitu kata "nol", "satu", "dua", "tiga", "empat". Struktur jaringan dari sistem neuro-fuzzy yang digunakan terdiri atas satu input layer, empat hidden layer dan satu output layer. Sistem ini telah diuji dengan membandingkan struktur neuro-fuzzy dengan neural network. Hasil yang dicapai memperlihatkan bahwa sistem neuro-fuzzy memberikan hasil yang lebih baik dibandingkan dengan sistem neural network. Waktu pembelajaran sistem neuro-fuzzy lebih cepat dibandingkan dengan neural network. Untuk 270 sampel suara, sistem neuro-fuzzy menyelesaikannya dengan 160.000 iterasi sedangkan neural network membutuhkan 500.000 iterasi. Nilai persentase kebenaran tertinggi dari sistem Neuro-fuzzy mencapai 96,36 % sedangkan sistem neural network saja mencapai 62,86 %. Kata kunci : neural network, neuro-fuzzy, pengenalan kata.

Published

2004-06-25