| In the field of smart home, researches have a lot of new methods for technical studies. While, for a whole system of smart home, there must be have not only hardware system but also software system. The system cannot ignore the personalized experiences and internal needs of the user. In order to a harmonious human-computer interaction, a software system has been studied, besides a hardware system. The intelligence is implemented by the emotion. And the virtual human can have intelligent interaction with the users by the emotional interaction. In the paper, for the purpose of intelligent, based on the smart home, affective computing is the main theory, and the human-computer interaction is main method. Through giving emotion to virtual human, the virtual human can interact with users, and the key problems and core technologies have been investigated.Contents and innovations are shown in the following:(1) The emotion of human can be expressed by dimensional space in the emotional dimensional theory in psychology. But how many dimensions are they very appropriate in affective computing? Researchers have not reached a consensus. For the problem of speech-to-text, this paper put forward a new method in human and computer interaction by a virtual human. A5points semantic differential scale has been proposed, which can estimate16emotional dimensions in psychology, with a designed questionnaire. Compared with the9points semantic differential scale, the new scale is more concise. The data derived from estimating20different texts by50testers, and were calculated by mean and variance, and has shown the validity. Based on the new scale, the16different emotional dimensions have been estimated by the same testers and other designed texts. From the product monument correlation and principal component analysis, the paper has some conclusions. That is, the16different emotional dimensions can be classified5classifications. And the two dimensions, the strength and the positive and negative properties, can describe70%emotion of human in speech-to-text. And others can contribute relatively fewer information. So, two dimensions of emotion can be expressed the main emotional information of virtual human and users. Three or four dimensions of emotion can also be used, if a higher standard is needed.(2) The emotion of human can be described by some basic feelings and complex feelings with discrete emotional space in psychology. And the most representation is the Hidden Markov Model (HMM). When virtual human has interaction with users, the HMM can be applied to describe the emotional states. But this model lacks of cognitive factors. For this problem, an improved model is proposed, which has the function of cognition and memory. And the cognitive factor is decrypted by the mood. By defining emotional energy of mood, the mood has closely related with the time, using Laplace transform and inverse change of the energy of mood. And according to physiochemical cycle, which is impacted on the mood, the variable model of mood can be built to make up the original model. At the same time, in order to increase mnemonic function, the emotional memories have been added to the original model. From the analysis of the data, the modified model can enhance the probability of the emotional value, which is larger. Meanwhile, it can also weaken the probability of the emotional value, which is smaller. The two modified functions have the coherence with our cognition.(3) In order to analysis the emotion in the speech-to-text by virtual human, a hierarchic method was first proposed based on emotional words database. The emotional words have many labels, and the database have extensional function. Then, in order to filter out the emotional words from many words, the tagged words have been selected as emotional words, by reducing the dimension of parts of speeches. Finally, the texts have been analyzed from the hierarchic aspect. And the three levels are words-sentences-texts. The emotional states of the speech-to-text can be analyzed by the above models, that is, the dimensional model with two dimensions and the discrete model with cognition and memories. In the first model, the strength and the positive and negative properties of the three levels are defined. The values of positive and negative properties can be showed by emotional words database. And the emotional strength can be calculated by formulae, which are suitable for common and adversative sentences. In the second model, the emotional analysis of the speech-to-text can be described by the modified HMM model, with researching impacted parameters in the variable mood model, and with designing parameters in the memorial model.(4) In order to achieve emotional interaction, an interactive system for the smart home is designed in the paper. In the system, virtual human can interact with users with emotion, through very many functions. As a virtual butler in system, virtual human can not only interact with users with emotion, by using the technology of speech interaction, but also implement some basic functions. Such as, the personal identification, the records of events, the financial analysis, the weather forecast, the web browsing and other basic functions. Meanwhile, the personalized functions have also designed in the system. The first is a questioning and answering system, with the matching mode of sentences. The inputs of speech-to-text, is divided into seven different types of sentences. And each sentence has different rules. Using these rules and modes, the inputs can be matched to the appropriate outputs. Through experiments, the matching mode of sentences can increase the accuracy in the interaction between virtual human and users. The second fuction is chatting system based on speech-to-text. The dimensional model and discrete model can be used to judge the emotional states of virtual human. In the results of the program, there are some similar properties with the two different models. And they can also achieve interaction between virtual human and users. |
PDF Full Text Request