Recognition and Classification of Emotions in Driver Voice Data Streams

Abstract

Voice-based emotion recognition presents a fundamental challenge in the field of automotive driving, as the driver's emotional state directly influences their cognitive abilities and decision-making processes. Traffic conditions, such as congestion, rush hour, road infrastructure failures, or accidents, often elicit a range of emotional responses. This study employs semi-synthetic data to analyze continuous audio streams, evaluating the competitiveness of the Adaptive Random Forest (ARF) algorithm for emotion detection in driving scenarios. The proposed methodology integrates a hybrid strategy of real and synthetic data to model dynamically evolving emotional patterns. The experimental results demonstrate that using a mixed data stream significantly improves model reliability and learning capacity, validating the effectiveness of ARF in emotion recognition tasks. These findings not only confirm the potential of the ARF algorithm in real-time emotion recognition systems but also provide valuable insights for the future development of intelligent transportation systems and non-intrusive driver monitoring systems