婴儿情绪信息的模式识别技术研究与实现

发布时间：2019-04-16 19:24

【摘要】：婴幼儿专家的研究成果表明,婴儿的情绪表达不仅是与外界交流的主要方式,而且是反映其生理和心理需求、心身健康状态乃至智力发育水平的重要信息来源。近年来,婴儿情绪信息的研究已经引起了人们极大的兴趣并成为相关领域正在探索之中的新兴前沿研究热点。从已有的研究工作来看,婴儿的语音信息是最便于准确采集并能体现婴儿独特的语言运动方式和情感表达特征的重要信息,受到了研究学者们的普遍关注。然而,目前尚缺乏统一的婴儿情绪分类方法及相应的语音信息特征描述,特别是对于蕴涵着丰富情绪信息的婴儿笑声和哭声,在其内涵意义的识别与理解上尚未形成统一的看法。本文在作者亲身体验的实践基础上,采用模式识别技术对上述问题作了探索性研究。首先,从婴儿的发生器官结构及其情绪表达特征入手,结合婴儿的主要生理与心理需求和其所处的环境特点,对婴儿的情绪状态分类与相关的语音情绪信息作了分析。然后,通过线性预测参数(LPC)、线性预测倒谱参数(LPCC)和Mel尺度倒谱参数(MFCC)等信号分析的技术参数,对婴儿语音情绪信息的数据采集和预处理过程及相应的特征参数提取方法进行了研究。在此基础上,本文进一步探讨和比较了人工神经网络(ANN)、隐马尔可夫模型(HMM)、动态时间规整(DTW)等方法应用于婴儿情绪信息模式识别的可行性。经过综合比较,本文采用了MFCC参数和DTW方法,针对婴儿最常见的高兴、饥饿、困倦三种典型的身心状态所表达的情绪信息作模式识别研究,并给出了其技术实现方法和实验测试结果,取得了良好的识别效果。本文的研究成果为相关领域的研究工作提供了重要的探索性启发。
[Abstract]:The research results of infant experts show that the emotional expression of infants is not only the main way to communicate with the outside world, but also an important source of information to reflect their physiological and psychological needs, psychosomatic health status and even the level of intelligence development. In recent years, the study of infant emotional information has aroused great interest and become a new frontier research focus in related fields. According to the previous research work, the infant's speech information is the most important information which is convenient to collect accurately and can reflect the infant's unique language movement and emotion expression characteristics, which has been paid more and more attention by the researchers. However, there is still a lack of a unified classification of infant emotions and the corresponding description of the characteristics of voice information, especially for infants with abundant emotional information, including laughter and crying. The recognition and understanding of its connotation and meaning have not yet formed a unified view. On the basis of the author's personal experience, this paper makes an exploratory study of the above problems by using pattern recognition technology. Firstly, the classification of infant's emotional state and the related phonological emotional information were analyzed based on the structure of the infant's generating organ and its emotional expression characteristics, combining the main physiological and psychological needs of the infant and the environmental characteristics of the infant. Then, through the linear prediction parameter (LPC), linear prediction Cepstrum parameter (LPCC) and Mel scale cepstrum parameter (MFCC), the technical parameters of signal analysis are analyzed. The data acquisition and pre-processing process of infant speech emotion information and the corresponding feature parameters extraction method were studied. On this basis, this paper further discusses and compares the feasibility of applying artificial neural network (ANN), hidden Markov model (HMM), dynamic time regularization (DTW) to infant emotional information pattern recognition. After a comprehensive comparison, this paper adopts MFCC parameters and DTW method to study the emotional information expressed in three typical physical and mental states of infants: happy, hungry and sleepy, and makes a pattern recognition study on the emotional information of the three typical physical and mental states of the infant, namely, happy, hungry and sleepy. The technical realization method and experimental test results are given, and good recognition results are obtained. The research results of this paper provide an important exploratory inspiration for the research work in related fields.
【学位授予单位】：复旦大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：R174;TN912.34

【参考文献】