Assessing the Impact of Feature Selection on Neural Network Performance in Audio Dialect Identification
Keywords
Loading...
Authors
Issue Date
2024-11-03
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
This study aimed to investigate the role of feature selection in the performance of neural networks for audio dialect identification, and to ascertain which features are most effective for capturing dialect-specific information across different dialects and phonemes. While previous research, such as that conducted by Tawaqal and Suyanto (2021), has demonstrated accuracy rates of over 80% with Mel-Frequency Cepstral Coefficients (MFCC), this study found less satisfactory results, with the highest accuracy score reaching 60%. The type of feature vectors that were significant were the number of frames around the vowel nucleus, the number of to be classified dialects and the different vowels. Non-significant type of feature vectors were parameters connected to the network (type, number of layers), inclusion of derivative and MFCC parameters.
Description
Citation
Supervisor
Faculty
Faculteit der Letteren