Speech Emotion Recognition using Supervised Bayes Learning on Digital Features of Multi-Label Data Corpus

Abstract

Emotion recognition from speech audio signal plays a critical part in several applications in this digital era of information systems include, including  many such as forensic services, operator or driver emotion monitoring in safety and security domain of large scale industries. Speech is a distinct human trait that is utilised to express and communicate person's point of view to others. Emotion recognition  from speech audio signals is the process of obtaining a presenter's emotions from the speech audio signal of respective presenter. Machine learning is a possible arena for developing emotion recognition systems, which is a critical research goal in current engineering research. Extraction of features, feature engineering, and classification are the three main phases in speech emotion recognition. Despite the fact that there are powerful machine learning-based emotion identification algorithms for speech audio signals, detection rate with maximum specificity and sensitivity is not scalable from most of these modern methods. The accuracy of detection is usually proportional to the features used to train the classifier. As a result, feature optimization for emotion recognition from speech audio signals is an obvious research goal. This article presented a unique technique called Supervised Bayes Learning on Digital Features (SBL-DF). The trials compared the suggested approach's performance to an earlier model  portrayed against a similar goal in order to scale its performance. The findings of an experimental study show that the proposed feature optimization strategy is possibly scalable. The benchmark classifiers' detection accuracy, with minimal false alarms.