Pronunciation Detection of Arabic Alphabet Phonemes Using AlexNet

Pronunciation detection is a crucial part of people’s daily lives, serving as a bridge for communication between languages using digital devices. Moreover, given the importance of Arabic in Indonesia for educational and religious purposes, individuals must learn this language with accurate pronunciation and phonemes to comprehend sentences correctly. To address this, research has been conducted to develop a system for detecting the pronunciation of Arabic alphabet phonemes using the AlexNet classification model. The pronunciation detection process involves converting sound files of arabic alphabet pronunciations into mel-spectrogram data, which is then used as input for training and testing the AlexNet classification model using the Atratified K-fold cross-validation algorithm. From the experiment, the phoneme classification accuracy without augmentation for the Amna Asif dataset is 84.7% and for the mixed data is 84.1%. While in the classification of phonemes with augmentation, the results obtained are 91.1% for Amna Asif dataset and 90.3% for mixed dataset.