Multi-level processing for continuous speech recognition in natural environment

Citation

Praveen, Edward James and Kit, Mun Hou and Aravind Vaithilingam, Aravind Vaithilingam and Alan Tan, Wee Chiat (2018) Multi-level processing for continuous speech recognition in natural environment. International Journal of Pure and Applied Mathematics, 119 (12f). pp. 15011-15024. ISSN 1311-8080, 1314-3395

[img] Text
122.pdf
Restricted to Repository staff only

Download (587kB)

Abstract

In a natural environment, the performance of a Hidden Markov Model (HMM) based speech recognition system degrades with noise. To overcome this limitation, additional processing techniques are required. This paper involves the design of sentence recognition system with multi-level processing techniques like Dynamic Time Warping (DTW) and Multi-Instance Training (MIT). This entire process involves two phases: training and recognition. The training phase of speech recognition involves preprocessing the signal to eliminate noise by band pass filtering, pre-emphasis, Voice Activity Detection (VAD), DTW and MIT Training. In the recognition phase the test sentences are processed into individual words 15011 and compared with training data and the recognized words are presented as text. Two stages of experiments are performed. In the first stage, a measure of accuracy called the Word Error Rate (WER) is calculated for the stand-alone system and estimated as 24.1. In the second stage, DTW is performed during preprocessing and MIT during training in a sequence of steps and the WER for each step was obtained as 23.6 and 23.4 respectively. The result shows that there is a 2.9% decrease in WER with preprocessing the basic system with DTW and MIT based training. The system has strong adherence to mathematical concepts and hence it is reliable, stable and suited for resource constrained environments

Item Type: Article
Uncontrolled Keywords: Speech recognition, feature extraction, Cepstral coefficients, Pre-emphasis, Estimation
Subjects: T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7800-8360 Electronics > TK7885-7895 Computer engineering. Computer hardware > TK7895.M4 Memory systems
Divisions: Faculty of Engineering and Technology (FET)
Depositing User: Ms Rosnani Abd Wahab
Date Deposited: 10 Mar 2021 20:11
Last Modified: 10 Mar 2021 20:11
URII: http://shdl.mmu.edu.my/id/eprint/7451

Downloads

Downloads per month over past year

View ItemEdit (login required)