Abstract
When society enters the era of informatization with the development of Internet information technology, people have been inseparable from the environment of artificial intelligence in work, study and even life. Nowadays, all kinds of appliances in life are oriented towards networking and intelligence, and the operation mode has also changed from touch buttons to more advanced control methods such as voice command control. Therefore, in terms of the requirements of the intelligent speech recognition in the family, according to deep learning technology, an acoustic model based on speech feature recognition and adopted DNN-HMM is designed in the paper, where the noise is filtered through the underlying network, and the high-level network is adopted to extract voice features. Moreover, through the improved algorithm, the linear unit is used to replace the nonlinear activation function so as to obtain better performance. Besides, after experimenting the research content in the paper, the deep belief network and the deep auto-encoder model are subjected to feature training, which proved that the speech recognition accuracy of the speech recognition system model has reached the design requirements. In other words, the research content in the paper meets the needs of home smart home voice recognition, and has made efforts to design smart homes that meet people’s needs.