Answers are grounded in this paper and cite their sources.
In this article, the authors present a long short-term memory (LSTM)-based approach for speech emotion recognition. First, they provide a detailed introduction to the LSTM model's fundamentals, along with information on how to calculate the input gate, forget gate, candidate cell state, and output gate. They then increased dataset by applying methods for enhancing audio data, such as pitch transformation, time shifting, adding noise, and time scaling, all of which strengthen the model's resilience and capacity for generalization. Lastly, the authors conducted a detailed analysis of this method by evaluating the Chinese Academy of Sciences Institute of Automation and Interactive Emotional Motion Capture datasets. The results show that the LSTM model performs well in identifying neutral, happy, sad, and angry emotions on the Institute of Automation and Interactive Emotional Motion Capture dataset, with accuracy rates of 0.875, 0.832, 0.805, and 0.850, respectively. On the Chinese Academy of Sciences Institute of Automation and Interactive Emotional Motion Capture datasets, the LSTM model also performs well, with accuracy rates of 0.885, 0.840, 0.822, and 0.863, respectively.
No takes yet. Share an insight, caveat, or question.
Tong et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: