Non-Speech Body Sounds的感知、识别与应用研究

发布时间：2018-03-29 07:51

本文选题：Non-Speech　切入点：Body　出处：《南京大学》2017年硕士论文

【摘要】：我们将人体内部发出的声音称为Non-Speech Body Sounds,包括呼吸声,吃饭声,喝水声,咳嗽声,睡眠声等一系列人体在日常生活中由身体内部产生的细微的声音。这些人体内部的一些声音往往能够传递出不少有关健康的信号。因此,Non-Speech Body Sounds的研究目前受到了学术界和工业界的关注。在学术界,越来越多的研究所开始关注利用可穿戴设备来收集Non-Speech Body Sounds进行人体相关状态和行为的研究,并且试图通过这样的研究能够一定程度上发现人体生理或心理的生物节律和隐藏疾病。在工业界,各种各样小巧灵敏的传感器为我们的研究提供了可能,利用这些传感器可以准确而又方便的发现各种人体内部声音,为我们的研究提供基础。目前的已有成果利用这些传感器来进行Non-Speech Body Sounds的感知,但这样的感知效果却不尽如人意。另外在这样的感知条件下,现有工作通过机器学习的方法进行识别,得到的结果准确率不高。基于这样的研究现状,本文旨在感知和识别Non-Speech Body Sounds,通过设计专有的可穿戴设备提高感知的效率同时通过时域、频域和梅尔倒谱系数的分析提高识别效果。本文通过设计专有的可穿戴项链感知Non-Speech Body Sounds,这套设备采用小巧的设备和合适的麦克风进行Non-Speech Body Sounds的感知,可以提高感知的效果。通过特有的技术框架进行数据处理分析,将感知到的Non-Speech Body Sounds进行时域、频域和梅尔倒谱系数的分析,然后利用帧维度和窗口维度进行特征的提取和选择,最后进行分类得到最终的结果。针对上述研究背景和解决思路,本文首先分析了 Non-Speech Body Sounds所涉及的具体研究问题,对研究现状进行了总结和分析,在此基础上,本文提出了基本思路:建立Non-Speech Body Sounds感知与识别系统结构总框架,包括硬件设计,系统设计和算法介绍,然后进行系统实现和评估,最后提出示范应用——抽烟行为检测应用,组织相关实验。本文的成果如下:1.本文设计一套基于智能项链的可穿戴设备,并基于此设备收集Non-Speech Body Sounds。该设备轻巧易于穿戴,可用于日常生活中;2.本文设计并实现了部分Non-Speech Body Sounds的感知和识别系统,其中包括吃东西,喝水,普通呼吸,深呼吸,叹气,咳嗽,吸鼻子的声音。通过可穿戴设备感知数据,我们的系统可以对数据进行层次化的处理并进行分类,并通过真实数据进行系统评估,结果展示我们的系统和前人的工作相比确实有较高的准确率,有较高的使用价值;3.本文实现Non-Speech Body Sounds的一个示范应用——抽烟行为检测来证明我们系统的价值。通过本系统收集人体内部非语音声音进行抽烟中深呼吸的判断,结合手势和打火机的声音来检测抽烟行为;4.本系统收集日常生活Non-Speech Body Sounds和若干种抽烟情况以及真实吸烟者的日常生活数据来组织实验,实验结果表明本系统有较高的准确率,可以在日常生活中应用。
[Abstract]:We will sound within the body called the Non-Speech Body Sounds, including breathing, eating, drinking water, cough, sound sleep and a series of human body in daily life from inside the body produce fine voice. Some of these voices inside the body are often able to pass a lot of health related signals. Therefore, study on Non-Speech Body Sounds at present by the academic circles and industry. In academia, more and more studies have started to pay attention to the use of wearable devices to collect Non-Speech Body Sounds of human state and behavior, and through such a study to human physiological or psychological biological rhythms and hidden disease found to a certain extent. In the industry, all kinds of sensitive sensors provide the possibility for us to study precisely and conveniently by using these sensors Find a variety of human internal voice, provide a basis for our research. The results of the current use of these sensors to Non-Speech Body Sounds perception, but the perceived effect is not satisfactory. Also in this sense under the condition of existing work by machine learning method for identification, the result accuracy is not high. Based on this research, this paper aims to perceive and identify the Non-Speech Body Sounds, through the design of proprietary wearable devices to improve sensing efficiency at the same time by time domain, frequency domain analysis and Mel cepstrum coefficients to improve the recognition performance. In this paper, through the design of proprietary wearable sensing Non-Speech Body Sounds necklace, this set of equipment with compact and suitable equipment Non-Speech Body Sounds microphone can improve the perception, perception of the effect. The data through the special technical framework Theoretical analysis, Non-Speech Body Sounds will be perceived in time domain, frequency domain analysis and Mel cepstral coefficients, extraction and selection and then use the frame window dimensions and dimensions characteristics, finally classification to get the final result. Based on the above research background and ideas, this paper first analyzes the Non-Speech Body Sounds relates to the specific research questions study on the present situation, summarized and analyzed, on this basis, this paper puts forward the basic idea: to establish Non-Speech Body Sounds perception and recognition system structure framework, including hardware design, system design and algorithm is introduced, then the system implementation and evaluation, and finally puts forward the demonstration application - smoking behavior detection application, organize the relevant experiment. This paper the results are as follows: 1. this paper designs a set of intelligent wearable devices based on the necklace, and based on the Non-Speech Body Sounds. collection equipment The light equipment easy to wear, can be used in daily life; the 2. part, this paper designs and implements Non-Speech Body Sounds perception and recognition system, including eating, drinking, normal breathing, deep breathing, sigh, cough, nose sucking sound. By wearing a very known data, our system can be carried out the hierarchical data processing and classification and evaluation system through the real data, the accuracy of the results show than do have higher system and our previous work, there is a higher value; the 3. Non-Speech Body Sounds a demonstration application - smoking behavior detection to prove our system by value. The system collects the internal body of deep breathing smoke non speech sound judgment, to detect the smoking behavior with a combination of gestures and lighter voice; 4. the system used in daily life Non-Sp Eech Body Sounds and several smoking cases and daily life data of real smokers are used to organize experiments. The experimental results show that the system has high accuracy and can be applied in daily life.

【学位授予单位】：南京大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP368.33;TP212.9

【相似文献】