Application Research of Speech Instruction Recognition in Human-computer Interaction

Abstract

Speech instruction recognition is an important research direction in the field of human and intelligent machine interaction. With the development of deep learning, more and more researchers pay attention to the automatic speech recognition technology based on convolutional neural network. By establishing an end-to-end speech recognition model based on deep neural network, taking logarithmic amplitude spectrum features as the network input, using connectionist temporal classifiers to carry out time sequence classification, and using convolutional neural network to deal with the correlation between frames, an end-to-end speech recognition system is implemented. On this basis, taking the speech interaction scene of intelligent home appliances in the living room as an example, the speech control instruction database of smart home appliances is constructed, and the application research of speech instruction recognition and data instruction generation is carried out. The method of transfer learning is adopted. We train a pre-training model with open source data set firstly, and then re-train the model with self-recorded speech control instruction database. The experimental results show that the proposed algorithm achieves 12.31% WER on public AISHELL-1 dataset and 2.1% WER on our self-recorded speech control instruction database. The speech instruction recognition system realized in this paper can quickly recognize speech instructions and convert them into formatted data instructions that can drive intelligent home appliances, thus improving the communication efficiency between human and machine.