Research On Old People's Chatting Robot Based On Deep Learning

Intelligent life has arrived,and the development of robots will become more and more prosperous.The functions implemented are becoming stronger and more complex.Looking at the world,various types of robots are emerging in all aspects of life.There are chat-type customer service robots.Robots that work tirelessly on the production line,and a very small number of high-end surgical robots,etc.,ranging from digging on the moon to deep-sea treasure hunting,they provide people with a variety of convenient services.The 14 th Five-Year Plan proposes to strengthen the social security of the elderly,put people first,and provide the elderly with personalized services to ensure the quality of life of the elderly in the complex situation of social diversification.The purpose is to empower the aging society in the future and relieve social pressure.Designing an accompanying nursing robot is a good entry point and research direction.Small talk and dialogue are the basic and most important skills of nursing robots.Considering that the thinking logic and the ability of verbal expression of the elderly are not as clear and coherently as the young people,in such a complicated situation,it is necessary to capture more accurate expression meanings.Traditional speech recognition and dialogue generation have been optimized and improved,and some additional personalized functions must be added,such as step counting,wake-up and alarm functions to assist the elderly.Companion chat robots include speech recognition and dialogue generation.The structure of the article also focuses on these two areas for research.In terms of speech recognition,we first studied the current mainstream end-to-end semantic recognition framework,and improved it for the elderly,and proposed an incomplete end-to-end recognition framework that incorporates prior knowledge to strengthen the ability to recognize incoherent speech.After understanding the principle of speech recognition and the complete recognition process,the limitations of the traditional acoustic model algorithm Hidden Markov Model(HMM)are analyzed and discussed,and the use of Connectionist Temporal Classification(CTC)to achieve speech is improved.As an acoustic model,fusion of different neural networks to achieve speech recognition and experimental analysis,it is found that the extraction of global features combined with convolutional neural network is better than recurrent neural network recognition.Regarding the dialogue generation part,we first analyzed the two mainstream frameworks of retrieval style and generative style.According to the characteristics of small chat,we selected the generative style with high response interest.The encoder-decoder-based Seq2 Seq model is an imported product from the translation system.The one-to-one correspondence is not very consistent in a dialogue of unequal length between one question and one answer.To this end,a fusion of the attention-based algorithm model Attention mechanism and seq2 seq model is proposed.Effectively alleviate the problems of inconsistency in the length of the conversation sequence and confusion in the expression of the sentences for the elderly.For further optimization,the response effect is better,and the extraction of keyword expressions is strengthened.At the same time,it does not appear in the corpus of the training model in the response results.New words.A seq2 seq model fused with a pointer generation network is proposed.Using automatic evaluation indicators and artificial indicators to evaluate the generated responses,it is found that the improved method is better than the previous method,and the output response results are more natural,diverse and interesting.Finally,the complete system is presented on the Android platform to accompany users to chat and relieve boredom while adding auxiliary functions to make the user experience more real and useful.