Використання методів машинного навчання для розпізнавання емоцій мовлення

Abstract

This research examines the performance of various pitch detection algorithms within the context of Speech Emotion Recognition, specifically focusing on accurately classifying human emotions from audio signals. Emphasis is placed on machine learning and signal processing methodologies designed to categorize emotional states effectively. Results indicate a considerable improvement in classification accuracy when combining both time-domain and frequency-domain features, with the highest-performing model surpassing 85% accuracy on popular datasets such as CREMA-D and RAVDESS. These findings underscore the promising capabilities of sophisticated SER models for enhancing human-computer interaction, while also emphasizing the importance of continued research toward developing robust and unbiased systems suitable for practical, real-world scenarios.