<![CDATA[This study develops a speech emotion recognition system for the Indonesian language using deep learning combined with hybrid audio features. A total of 300 speech samples from the IndoWaveSentiment dataset, labeled into five emotion classes neutral, happy, surprised, disgusted, and disappointed were used. The extracted features include Mel-Frequency Cepstral Coefficients (MFCC), Hilbert Multispectrum, and Cochleagram, capturing frequency, amplitude, and auditory perception characteristics of the speech signals. Three deep learning models were implemented: Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and a hybrid CNN-LSTM model. All models were trained and evaluated using accuracy, precision, recall, F1-score, and Area Under the Curve (AUC). The CNN model achieved the highest validation accuracy of 98.3%, followed by LSTM (91.6%) and the hybrid CNN-LSTM (85%). All models demonstrated strong reliability with AUC values above 0.9. A local web-based interface using Flask was developed to enable real-time audio recording or upload for emotion prediction. This research contributes to speech emotion recognition and applied informatics in the Indonesian language, demonstrating that hybrid audio features combined with CNN provide accurate and practical solutions for emotion classification.]]>
Copyrights © 2026
