A Novel Emotion Recognition Model based on Speech Processing

During the past two decades, recognizing emotions in people's voices (known as recognition of speech emotions, or SER) has been a hot issue in the field of language modeling. Speech being the major communications channel, the speech emotion recognition system has become an integral aspect of the HCI area in order to comprehend the emotional state of humans from speech and respond appropriately. Sentiment analysis and emotion recognition in spoken language is a complex area of study in human multimodal communication. Yet, there are circumstances where just audio data is available, such as phone conversations. In this research, we offer a unique approach to speech emotion detection based on attribute creation and hybrid learning that can handle a limited amount of emotion data. To begin, the voice signal's acoustic characteristics are extracted and then merged to generate several novel feature sets. Decision Tree, Linear SVC, CatBoost, and AdaBoost are each single weak learners, and their respective weights are learned dynamically using the softmax predictive model. Finally, an ensemble classification method called DSCA is built that combines the aforementioned four separate weak learners using the weighted mean probability voting technique and the automated weights gained throughout the training process.