Unveiling Emotions: Long-Short Term for Speech Emotion Recognition

Speech Emotion Recognition (SER) research has traditionally concentrated on analyzing the native languages of speakers, primarily emphasizing European and Asian languages. This study explores speech emotion recognition in Hindi emotion classification. This research employs advanced machine learning and deep learning techniques on diverse Hindi emotional expressions. Features like MFCCs are used, and LSTM algorithms, including SVM, CNN, and RNN, are compared for result analysis. Results show the system's effectiveness in accurately classifying emotions, providing valuable insights into Hindi emotional patterns, and enabling applications in affective computing and human-computer interaction. The research presents an innovative approach to developing a curriculum for machine learning, focusing on training deep neural networks for speech emotion recognition. The primary challenge addressed is establishing the difficulty order of training data, which is tackled by utilizing the level of disagreement among human evaluators as an indicator of difficulty. Through leveraging this disagreement metric, the proposed method shows a notable enhancement in model performance compared to baseline methods. This improvement underscores the effectiveness of the approach in elevating the accuracy of speech emotion recognition. The accuracy we have gotten by implementing the LSTM model on the SER system is 94%. In comparison with other models, the accuracy is more of LSTM model.