Value leadership of large language models based on speech recognition and emotion classification algorithms

An integrated algorithm was proposed to better address the issues of speech recognition and emotion classification in the natural language processing. This algorithm was committed to accurately converting speech information into text form and conducting sentiment analysis on it. The simulation experiment results showed that the loss value of the model in the training set was about 0.35, and the loss value in the validation set was about 0.99. After feature extraction, the accuracy, gain rate, echo value, and F1 value of the model were improved to 0.87, 0.88, 0.88, and 0.88, respectively, showing significant improvement. Compared with other similar models, the proposed model had a higher overall recognition rate, especially in emotions such as anger (90.25%), fear (89.78%), and disgust (90.11%). The above results show that this model can better understand and generate emotional language expressions and provide better services for natural language understanding.