基于跳层连接卷积网络的语音情感识别研究

With the development of intelligent devices, voice recognition technology, as one of the most convenient human-computer interaction technologies, has an increasing demand. Some products, such as voice intelligent assistants, have become commercial. However, according to the general reflection of users, the current major voice assistants are far from natural and smooth communication and interaction, because voice assistants lack emotional expression. With the deepening of the research on human-computer interaction technology, how to recognize the emotional information in speech signals has become one of the urgent problems to be solved. At present, the academic research on speech emotion recognition only stays on the use of traditional deep learning model. In order to study the deep learning network more suitable for speech emotion, this paper introduces the layer hopping connection into the convolution neural network, uses the spectrogram as the input feature, uses the two-way LSTM for time series modeling, systematically analyzes the impact of introducing the layer hopping connection on emotion recognition after different convolution layers of the convolution neural network, and selects the neural network model with the best performance, Finally, 1.06% performance improvement is achieved.