A Novel Approach to Audio-Based Emotion Recognition Using Transfer Learning Algorithm (Wav2Vec2.0)

Aim: This study aims to increase the precision and reliability of the real-time application using Wav2Vec 2.0 for audio data emotion recognition. Materials and Methods: There are two groups in this study. Group 1 is the previous self-supervised learning model with 26 samples, while Group 2 is the Wav2Vec 2.0 proposed model with 26 samples. G Power value at 80 %, threshold 0.05%, and 95 % confidence level. Mean accuracy, precision, variance, and standard deviation will be the used metrics in this performance evaluation. Result: Therefore, it was more accurate than the self-supervised model with an accuracy of 97.3 % and a precision of 97.8 %, respectively. Conclusion: This accuracy and consistency make the Wav2Vec 2.0 considered quite effective, even in other domains such as psychology, entertainment, and customer service, based on the prior-mentioned research.