SVM BASED SPOKEN NUMERALS RECOGNITION FOR PUNJABI LANGUAGE by Satbir Kaur
Abstract
The advances in digital signal processing technology had led the introduction of speech processing in several applications like speech compression, enhancement, synthesis, and recognition. In this thesis the problem of speech recognition is studied and a speaker dependent, medium vocabulary, isolated numerals speech recognition system has been developed for Punjabi language. The study implements the Support Vector Machine based isolated numerals speech recognizer in three steps. The primary step performs the end point detection, framing and windowing of the speech signal. The second step includes speech feature extraction using Mel Frequency Cepstral Coefficients. Finally, in the last step the Support Vector Machine is primarily used as recognizer and classifier for numeral of Punjabi language. The system is ready to understand the spoken utterances in Punjabi language, by removing the unwanted noise and unsuitable data from the speech signal and then translating the speech wave into quality of feature vectors. These vectors are given to SVM to predict the spoken word in written form. The Support vector machine classifier is wide used and given high accuracy. SVM has the ability to deal with high dimensional knowledge and is used in many real world problems like text categorization, character recognition and classification.
