婴儿需求表达语音信息的智能识别技术研究

发布时间：2018-03-18 09:05

本文选题：人机交互　切入点：情感需求　出处：《复旦大学》2014年硕士论文　论文类型：学位论文

【摘要】：随着科技的迅猛发展,人们对人机交互的能力要求越来越高,对于计算机如何达到拟人化,其中最重要的一点就是要对人的情感信息做到准确的智能识别。语音识别属于人机交互其中一项重要技术能力,并且已经成为国内外许多学者正在探索研究的新兴热点问题,然而传统的语音识别系统一般只反映了部分的信息,而忽略了语音中包含的情感信息,这样也降低了语音识别的准确率。本文将着重于对情感的声学特征进行研究,即通过对包含在语音中的说话人的情感特征的分析来识别说话人的情感信息,这也成为提高语音识别系统的识别率的重要指标。本文在基于作者亲身体验的基础上,采用模式识别方法对上述问题作了一些基于提高识别准确率的研究。首先,根据国内外学者对情感分类的研究,结合婴儿的心理、生理需求和所处的环境因素,将婴儿的情感需求进行分类定义。其次,分析目前婴儿情感语音识别的研究现状与实现技术,分析当前各特征参数提取的区别和意义。情感特征参数包括例如基音、能量等参数,并且分析每一个特征参数和婴儿情感特征的相关性。再次,通过分析语音识别的模式理论,包括K-近邻方法(KNN)、隐马尔可夫模型法(HMM)、神经网络的方法等模式识别理论对情感的模式识别可行性,提出KNN算法结合相应情感特征参数贡献率对情感语音识别的有效性。经过综合比较,本文将综合多维情感特征参数和KNN算法对婴儿的情感需求进行模式识别实验,并给出相应的技术实现方法以及实验结果。
[Abstract]:With the rapid development of science and technology, people have higher and higher requirements for the ability of human-computer interaction. Among them, the most important point is to achieve accurate intelligent recognition of human emotional information. Speech recognition belongs to one of the important technical capabilities of human-computer interaction, and has become a new hot issue that many scholars at home and abroad are exploring and studying. However, the traditional speech recognition system generally only reflects part of the information, but ignores the emotional information contained in the speech, which also reduces the accuracy of speech recognition. This paper will focus on the acoustic characteristics of emotion. That is to say, the speaker's emotional information can be recognized by analyzing the emotional characteristics of the speaker included in the speech, which has become an important index to improve the recognition rate of the speech recognition system. This paper is based on the author's personal experience. Based on the research of improving the accuracy of recognition, the pattern recognition method is used to study the above problems. Firstly, according to the research of emotion classification by domestic and foreign scholars, combining the psychological and physiological needs of infants and the environmental factors, The emotional needs of infants are classified and defined. Secondly, the current research status and implementation techniques of emotional speech recognition are analyzed, and the differences and significance of extracting each feature parameter are analyzed. The emotional feature parameters include, for example, pitch, speech recognition, speech recognition, speech recognition, speech recognition, speech recognition, speech recognition, speech recognition and speech recognition. Energy and other parameters, and analyze the correlation between each characteristic parameter and the emotional characteristics of the baby. Thirdly, by analyzing the pattern theory of speech recognition, Including K-nearest neighbor method (KNNN), hidden Markov model (HMM), neural network (NN), and so on. This paper presents the validity of KNN algorithm combined with the contribution rate of corresponding emotional feature parameters to emotional speech recognition. After a comprehensive comparison, the multi-dimensional emotional feature parameters and KNN algorithm are combined to carry out pattern recognition experiments on the emotional needs of infants. The corresponding technical realization method and experimental results are also given.
【学位授予单位】：复旦大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN912.34

【参考文献】