Robust isolated word speech recognition system
Rajan Sobhana Rashobh
Date of Issue2008
School of Electrical and Electronic Engineering
The aim of this study is to develop a Robust Isolated Word Speech Recognition System. A Dynamic Time Warping based Automatic Speech Recognition system is implemented which can recognize digits zero through nine, pronounced in English by different speakers under different conditions. Pattern matching method is adopted here. This system can be used for applications such as phone dialing. Mel Frequency Cepstral Coefficients (MFCCs) are used as the feature vectors in this project. The templates are created by using the combination of Dynamic Time Warping and Vector Quantization techniques. Templates are created with different sizes and the results are evaluated. The recognizer is evaluated against different noise conditions and speakers. The evaluated results are discussed and finally, this study also put forward the recommendation and future works to extend this project.
DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing