Speech Emotion Recognition Using GRU and LSTM

Speech emotion recognition (SER) is a hot topic in research, with applications in human-computer interaction, mental health screening, and customer service. In this research, our objective is to develop an accurate and reliable SER system capable of emotion recognition in speech data. We address important issues such as feature extraction, model selection, data imbalance, robustness, and real-time inference using a labeled dataset of various emotional expressions in voice. To extract complicated shapes from auditory characteristics, our method combines deep learning techniques that include long-short-term memory (LSTM) and gated recurrent units (GRU). We carefully manage the data imbalance to ensure that emotions are fairly represented in the training set, and with the implementation of our functional LSTM and GRU models in the training set, we were able to achieve the precision of 62 % and 90 %, respectively.