FPGA Implementation of Speech Recognizer for Isolated Words

This paper investigates the implementation of feature extraction of pre-recorded speech signal on FPGA board. The technique used is Mel Frequency Cepstral feature extraction. This is followed by the HLS implementation of comparison of feature vectors with the existing data. Comparison of vectors is done by Dynamic Time Warping (DTW) method. The aim of this paper is to implement the above techniques on FPGA and compare the results of the output with that of outputs from software. Also utilization of resources on FPGA board are investigated to ensure feasibility of implementation. In near future, the speaking with machines would become most common in everyday lives. This paper would serve a major part in speech recognition on hardware which is most preferred over the software implementation due to the computational speed and security.