基于性别判定的声纹辨别与确认系统设计与实现

发布时间：2018-05-18 02:01

本文选题：性别判定 + 说话人鉴别　；参考：《电子科技大学》2014年硕士论文

【摘要】：在目前的刑事案件处理中,多是警务人员参与或者证人指认的方式来处理对犯罪嫌疑人的指认,这种方式无论从时间上还是从正确度上都是滞后的、不准的,不仅效率、正确率低,严重的甚至可能误导延后对嫌疑人的判定,导致更严重的后果。根据上述情况,本文将每个人的声纹特征作为个人特征标识存储于计算机中,在已获取到犯罪份子语音数据的情况下,通过本系统对目标说话人在众多已有的模型数据中进行声纹的鉴别与确认,将计算机作为一个快速鉴别罪犯的安全管家,做警务以及司法快速过滤以及确认嫌疑人范围的有效帮手。本文提出基于性别判定的与文本无关的说话人鉴别与确认相结合的方式,对嫌疑人身份进行确认。在默认已知嫌疑人所在语音库的情况下,系统首先识别出说话人性别,再根据性别分类结果进行说话人鉴别过程,得到与目标说话人最大相似概率的嫌疑人模型,最后进入说话人确认阶段近一步确定当前说话人与结果嫌疑人的相似度,决定是否“接受”该嫌疑人的身份,从而实现对嫌疑人的快速确认,加速办案效率,加快排除危险,保证人民的人身财产安全。本系统由五个子模块构建而成,分别为声纹采集模块,声纹训练模块,声纹认证模块,认证记录库管理模块,用户声纹管理模块,其中声纹认证模块又包括了性别判定、声纹辨认、声纹确认三个方面。这五个模块关联起来共同完成本系统的职能。本系统整体是Java Web结构,采用Play框架做Web服务器的开发,使用Mysql数据库存储所有用户的声纹状态以及嫌疑人比对的结果记录,使用麦克风进行语音采集,利用Matlab处理特征及模型。在系统设计时,需求与设计思路的描述均采用UML语言进行详细的阐述。在系统实现上,利用测试驱动开发的技术思想,在实际开发过程中先设计测试用例,再编写代码,边开发边测试。这样保证了代码的可扩展性、可维护、可复用性,确保开发能高效进行。
[Abstract]:In the current handling of criminal cases, most of them are involved in police personnel or the identification of witnesses to deal with the identification of criminal suspects. This method is lagging behind in terms of time and accuracy, not only is it not efficient, Low accuracy, serious or even misguided judgment of suspects can lead to more serious consequences. Based on the above situation, this paper stores each person's voiceprint feature as a personal signature in a computer, where criminal voice data have been obtained. Through this system, the target speaker is identified and confirmed in many existing model data, and the computer is regarded as a safe housekeeper to identify criminals quickly. Effective helper for policing and judicial quick filtering and for identifying suspects. This paper proposes a method of text independent speaker identification and confirmation based on gender judgment to confirm the identity of the suspect. Under the condition that the speaker is known by default, the system first recognizes the speaker's gender, then carries on the speaker identification process according to the gender classification result, obtains the suspect model with the maximum similarity probability with the target speaker. Finally, we enter the speaker confirmation stage to determine the similarity between the current speaker and the result suspect, and decide whether to "accept" the suspect's identity, thus realizing the rapid identification of the suspect, speeding up the efficiency of handling the case, and speeding up the removal of danger. To ensure the safety of the people's life and property. The system is constructed by five sub-modules, which are voice pattern acquisition module, voice pattern training module, sound pattern authentication module, authentication record database management module, user voiceprint management module, among which the voice pattern authentication module includes gender determination. Voiceprint recognition, voiceprint confirmation in three aspects. These five modules are connected together to complete the functions of the system. The whole system is Java Web structure, Play frame is used to develop Web server, Mysql database is used to store the voiceprint status of all users and record the results of suspect alignment, and microphone is used for voice acquisition. Matlab is used to deal with features and models. In the system design, the requirements and design ideas are described in UML language. In the realization of the system, using the technology idea of test-driven development, the test cases are designed in the actual development process, then the code is written, and the test is developed while developing. This ensures the extensibility, maintainability and reusability of the code, and ensures that the development can be carried out efficiently.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN912.34;TP311.52

【参考文献】