Research on speech recognition algorithm for power system operations based on convolutional neural networks

In the field of the deep integration of information and communication technology and artificial intelligence technology, the application of voice control technology in power system dispatching is very important. In order to solve the problem of voice command recognition in power operation and improve the recognition accuracy and stability, this study proposes a voice recognition method for power operation based on deep neural network algorithm. In the preprocessing stage, the method uses high fidelity microphone to collect voice signals, improves the signal quality with the help of noise reduction, filtering, data conversion and other technologies, and uses weighted finite state machine (WFST) and beam search algorithm to optimize the processing flow. In the acoustic feature extraction process, Mel frequency cepstrum coefficient (MFCC) is used to simulate the perceptual characteristics of human auditory system on sound frequency, and linear discriminant analysis (LDA) is used to reduce the dimension of features, so as to enhance the computational efficiency and robustness of the model. The recognition algorithm uses convolutional neural network (CNN) to adaptively convert the speech signal, and outputs the recognition result combined with softmax function.