Central AsianResearch Index

SUPERVISED PERSIAN SPEECH EMOTION RECOGNITION USING DISTILHUBERT AND OPTUNA TUNING

Speech emotion recognition (SER) is recognized as a growing field with important applications in human computer interaction. In this study, attention is given to SER for the Persian language, which is regarded as a relatively underexplored domain. A lightweight ar- chitecture is introduced, in which a frozen DistilHuBERT model is used for feature extraction, and a bidirectional gated recurrent unit (Bi-GRU) block with a dense classification head is employed. Optuna is used for hyperparameter optimization, in which the Bi-GRUs struc- ture, learning rate, dropout rate, L2 regularization, and optimizer choice are tuned so that a balance between simplicity and performance is achieved. The proposed model is trained and evaluated with five fold cross validation, and an acceptable accuracy is obtained. These results show that competitive performance is achieved with a compact and optimized model in comparison with larger architectures.

Сведения

Опубликовано
2026 · Т. 14, № 1, С. 4-16
Язык
Английский
ISSN
2306-6172 · 2308-9822
Издатель
L. N. Gumilyov Eurasian National University
Аффилиации
Islamic Azad University