PENGENALAN TUTUR KATA TERISOLASI MENGGUNAKAN MFCC DAN ANFIS
Automatic Speech Recognition has achieved substantial success in the past few decades but more studies are needed because none of the current methods are fast and precise enough to be comparable with human recognition abilities. In the area of voice signal processing, neural network system has been...
| Main Authors: | , |
|---|---|
| Format: | Thesis |
| Published: |
[Yogyakarta] : Universitas Gadjah Mada
2013
|
| Subjects: |
| Summary: | Automatic Speech Recognition has achieved substantial success in the past
few decades but more studies are needed because none of the current methods are
fast and precise enough to be comparable with human recognition abilities. In the
area of voice signal processing, neural network system has been widely applied to
speech recognition process. Implementation of neuro-fuzzy system that combine
fuzzy system with neural network is possible to provide good results.
In this study the Adaptive Neuro Fuzzy Inference System (ANFIS) is used
for speech recognition in Indonesian with feature extraction using Mel-frequency
cepstrum algorithm Coefficient (MFCC). Neuro-fuzzy system has two initial
phases. The first phase is the system identification using Fuzzy C-Means (FCM)
algorithm to identify and initialize the parameters of the rules used in fuzzy
inference system (FIS). The second phase is fuzzy systems training using ANFIS
to optimize the parameters of fuzzy system. The system implements variation of
the number of fuzzy rule. System performance was evaluated using speech
samples for training and testing.
The test results using training data show that the highest average accuracy
of recognition is 100%. Testing with test data yield the highest average accuracy
rate of 89,33% with the number of fuzzy rules of 100. |
|---|