Analysis of Emotion Recognition System Based on Hybrid Neural Network

In order to enhance the resilience and generalisation capacity of the system in the context of minor alterations or noise contamination in the input natural language data, a personalised natural language emotion recognition system based on a hybrid network has been developed. A natural language acquisition module, based on a field-programmable gate array, has been designed for the purpose of collecting personalised natural language information. The preprocessing module facilitates the processing of the collected personalised natural language information, resulting in the generation of a log-Mel spectrum. The emotion recognition module employs a hybrid temporal network comprising a dynamic convolutional temporal network and a short-short-memory network. The speech information features are extracted from the Logmel spectrum through a dynamic convolutional neural network, while the text information features are extracted from the natural language text through a short-short-memory neural network. The speech and text information features are then fused through a fully connected neural network to output personalised emotion recognition results. The system has been demonstrated to be effective in the collection of personalised natural language information, the extraction of speech information features and the completion of personalised natural language emotion recognition. Furthermore, the system has been shown to be capable of maintaining high levels of accuracy in emotion recognition even when noise is present.