Research Article
Automatic Speech Recognition for Indian Accent Lectures contents using End-to-End Speech Recognition model
@INPROCEEDINGS{10.4108/eai.7-12-2021.2314531, author={Ashok Kumar L and Karthika Renuka D and Raajkumar G}, title={Automatic Speech Recognition for Indian Accent Lectures contents using End-to-End Speech Recognition model}, proceedings={Proceedings of the First International Conference on Combinatorial and Optimization, ICCAP 2021, December 7-8 2021, Chennai, India}, publisher={EAI}, proceedings_a={ICCAP}, year={2021}, month={12}, keywords={automatic speech recognition (asr) indian accent word error rate (wer) nptel lecture audio listen attend and spell (las)}, doi={10.4108/eai.7-12-2021.2314531} }- Ashok Kumar L
Karthika Renuka D
Raajkumar G
Year: 2021
Automatic Speech Recognition for Indian Accent Lectures contents using End-to-End Speech Recognition model
ICCAP
EAI
DOI: 10.4108/eai.7-12-2021.2314531
Abstract
In a variety of voice search applications, Automatic speech recognition (ASR) systems are used. The process of turning speech to text is known as automatic speech recognition (ASR). Most of the ASR research is happening using American and British accent. Hence, in this work we have made an attempt to convert Indian accent speech to text using NPTEL lecture audio. The proposed work involves speech to text using deep learning models for Indian accent speech. LAS has two main components one is based on sequence-to-sequence framework with a pyramid structure, by reducing the encoder steps in number the decoder must attend through the end-to-end process. The result obtained from the proposed work improve Word Error Rate of 14%.
Copyright © 2021–2024 EAI
