Accent Recognition Using Mel-Frequency Cepstral Coefficients and Convolutional Neural Network

DOI

10.2991/aer.k.211129.010How to use a DOI?

Keywords

accent recognition; speech recognition; MFCC; CNN

Abstract

Everyone has a different accent, the environment and culture can influence the difference in accents. Utilization of the recognition of the speaker’s accent can be used as a method to detect the speaker’s country of origin. Accent recognition belongs to the field of Automatic Speech Recognition (ASR), research on accent recognition is a step towards a smarter and more sophisticated ASR. Recently, ASR has become a trend in technology, such as virtual assistants. This study aimed to classify accents from several countries, namely English, Arabic, French, Spanish, and Mandarin, where all speakers used the same English Script. Previous research on accent recognition achieved 48.24 % using Mel Frequency Cepstral Coefficients (MFCC) and 2-layer Convolutional Neural Network (CNN). This study increases the accuracy by improving the pre-processing and the CNN model, the methods resulting 51.96 % accuracy using the similar of datasets as the previous study namely 1231 speakers and the methods namely Mel Frequency Cepstral Coefficients (MFCC) and 2-layer Convolutional Neural Network (CNN). By splitting the audio segment at the pre-processing and improving the model of CNN, it turns out to produce better accuracy.

Copyright

© 2021 The Authors. Published by Atlantis Press International B.V.

Open Access

This is an open access article under the CC BY-NC license.

Download article (PDF)

TY  - CONF
AU  - Dwi Sari Widyowaty
AU  - Andi Sunyoto
AU  - Hanif Al Fatta
PY  - 2021
DA  - 2021/11/30
TI  - Accent Recognition Using Mel-Frequency Cepstral Coefficients and Convolutional Neural Network
BT  - Proceedings of the International Conference on Innovation in Science and Technology (ICIST 2020)
PB  - Atlantis Press
SP  - 43
EP  - 46
SN  - 2352-5401
UR  - https://doi.org/10.2991/aer.k.211129.010
DO  - 10.2991/aer.k.211129.010
ID  - Widyowaty2021
ER  -