Research on Speech Information to Sign Language Translation Based on 1D-GoogLeNet and LSTM

In order to solve the problem of obtaining speech information for the deaf and mute population, a deep learning based intelligent speech to sign language translation technology is proposed. Firstly, a one-dimensional Google LeNet combined with Hidden Markov Model (1D Google LeNet HMM) is introduced on the basis of Gaussian Mixture Model Hidden Markov Model (GMM-HMM) to address the shortcomings of traditional speech recognition. In addition, a sliding window attention mechanism is introduced to overcome the speech recognition decoding problem of end-to-end connectionist temporal classification (CTC) algorithm. Finally, the study used long short-term memory network encoding and decoding to achieve the conversion of speech to sign language text. In the comparison of speech recognition performance, the average accuracy of the research model's speech recognition is 97.82%, and the stability is 98.25%, which is the best performance among similar models. In the comparison of end-to-end translation, the research model has the shortest conversion time and a conversion accuracy of 98.25%, performing the best among similar technologies. From this, it can be seen that the application effect of research technology is good, and the research technology will provide translation technology support for communication and exchange among deaf mute patients.