Mathematical Profile of Automatic Speech Recognition Algorithm

Abstract

This work provide mathematical insight to Automatic Speech Recognition (ASR) system’s algorithm such that, the intricacy of the system becomes a simplified correlation of the ASR algorithm to the physical form using the mathematical flowchart which clearly and uniquely show the link from one stage of the algorithm to the other. The mathematical profile of the ASR algorithm starts from the data input module, through noise cancellation module, voice activity detection module, pre-processing module, Linear Predictive Coding (LPC) based feature extraction module, then provides alternate root for both Dynamic Time Wapping (DTW) and Hidden Markov Model (HMM) based pattern matching module after which the output is fed to the final decision module of the ASR algorithm. The modern research outputs has improved the robustness of each stage of the algorithm but the approach used here focused on the basics of each stage which helps in easy and better understanding of the ASR system. It also aid in the evaluation and create necessary intuition for decoding problems of the recent ASR systems for new researchers in the research area.