Structural optimization of deep belief network theorem for classification in speech recognition

Speech is natural verbal form of communication in human beings. Each spoken word consists of phonetic combinations of vowels and consonants. Speech recognition is an application of the field of study in pattern recognition, applying pattern matching to phonetic patterns for identification of various linguistic objects including parts of speech. The various techniques to approach the model of this study follows in speech recognition are traditionally recorded the speech, extracts the feature from the signal, analyze the signal using Fast Fourier Transform (FFT) from the time series data set of speech and classifying the model using Deep Belief Network (DBN). DBN, itself has many advantages like feature extraction and classification that are used in several applications especially in image processing and signal processing. The aim of this study is to construct semi-automated feature representation that can improve the machine learning application model especially in speech recognition. The performance of DBN in accuracy of data classification, depends on the structure of DBN. This study uses a structure optimization of DBN which based on combined technique of evolutionary computation. The result of the experimental in structural optimization of DBN indicates the structure have an improvement of 100% on the simple traditional dataset.