| Issue |
EPJ Web Conf.
Volume 374, 2026
1st International Conference on Electronic, Optical Devices and Intelligent Systems (ICEODIS 2026)
|
|
|---|---|---|
| Article Number | 02002 | |
| Number of page(s) | 7 | |
| Section | Artificial Intelligence and Data Science | |
| DOI | https://doi.org/10.1051/epjconf/202637402002 | |
| Published online | 24 June 2026 | |
https://doi.org/10.1051/epjconf/202637402002
Fine-Tuning VGG19 with Mel Spectrograms for Amazigh Spoken Digit Recognition
1
Laboratory of Engineering Sciences, USMBA, Taza, Morocco
2
SDIA Team, UMP, Nador, Morocco
3
Laboratory of Applied Mathematics and Information Systems, UMP, Nador, Morocco
4
Laboratory of Optics, Information Processing, Mechanics, Energetics and Electronics, UMI, Meknes, Morocco
* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.
Published online: 24 June 2026
Abstract
This paper shows an improvement in speech recognition performance for the Amazigh language using the VGG19 model. We utilized a database of the first ten spoken Amazigh numbers, and we employed mel spectrograms as the visual representation of the audio. In the initial experiment, we froze the feature extraction layers of the VGG19, and trained only the custom classification layers, under the same training conditions as our previous work. This approach resulted in a recognition accuracy of 75.55%. Then, fine-tuning the entire model over 10 epochs improved the recognition accuracy to 89.33%.
© The Authors, published by EDP Sciences, 2026
This is an Open Access article distributed under the terms of the Creative Commons Attribution License 4.0, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.