Automatic Speech Recognition by Using Neural Network Based on Mel Frequency Cepstral Coefficient

Muhammad Daud  Abbasi; Zubair  Sajid; Shahzad Karim  Khawer; Syed Zain  Mir; Abdul  Basit; Muhammad  Kashif

doi:10.62019/vs3esy64

Authors

Muhammad Daud Abbasi Faculty of Engineering Science and Technology, Iqra University, Karachi, 75500, Pakistan
Zubair Sajid Faculty of Engineering Science and Technology, Iqra University, Karachi, 75500, Pakistan
Shahzad Karim Khawer Faculty of Engineering Science and Technology, Iqra University, Karachi, 75500, Pakistan
Syed Zain Mir Faculty of Engineering Science and Technology, Iqra University, Karachi, 75500, Pakistan
Abdul Basit Faculty of Computer and Information Technology, Indus University, Karachi, Pakistan.
Muhammad Kashif Faculty of Computer and Information Technology, Indus University, Karachi, Pakistan.

DOI:

https://doi.org/10.62019/vs3esy64

Keywords:

Speech Recognition, Short Term Energy, Zero Crossing Rate, Mel frequency Cepstral coefficient, Neural Networks, Feed forward Neural Networks, Back Propagation

Abstract

This paper deliberated and estimated the Neural Networks Automatic Speech Recognition (ASR) system based on an isolated small vocabulary speaker-independent manual cropping technique, from the training stage to the recognition stage. Besides this, the paper also examines three distinct blocks of speech recognition, i.e., Speech Preprocessor, Feature Extractor, and a Recognizer. Speech preprocessing involves windowing, framing, Short Term and Zero Crossing threshold energy, and End Point Detection calculation. Mel Frequency Cepstral Coefficients (MFCC) are extracted to represent the speech signal in frames and then passed through a Mel frequency filter. Multi-layer feed-forward network trained by the back-propagation method.