Penerapan Learning Vector Quantization (LVQ) dan Ekstraksi Ciri Menggunakan Mel-Frequency Cepstrum Coeffecients (MFCC) untuk Transkripsi Suara ke Teks

Abstract

Speech recognition by a computer is not an easy thing to do. Speech to text transcription is a technique that allows a computer to accept input in the form of spoken words and convert it into text. The purpose of this study is to model the neural network namely Learning Vector Quantization (LVQ) for speech to text transcription and determine the accuracy of speech recognition using MFCC feature extraction. The experiments are conducted by recognizing each syllable of the test data. The results show that the highest accuracy is 98.57% when the epoch value is 90, learning rate is 0.007, and learning rate decrement factor is 0.977. This accuracy is obtained by using the following MFCC parameters: sampling rate 11000 Hz, time frame 23.27 ms, overlap 0.39, and cepstral coefficients 13.