驾驶舱话音记录器音源识别和说话人识别的基础研究

项目名称： 驾驶舱话音记录器音源识别和说话人识别的基础研究

项目编号： No.U1233131

项目类型： 联合基金项目

立项/批准年度： 2013

项目学科： 电子学与信息系统

项目作者： 杨琳

作者单位： 中国民航科学技术研究院

项目金额： 43万元

中文摘要： 现代航空器上安装的驾驶舱话音记录器（Cockpit Voice Recorder，CVR）是必不可少的机载设备，它通过四个独立声道记录着驾驶舱内最近2小时或30分钟的各种声音，包括机组内话、陆空管制通话、音响警告等，分析CVR上记录的话语声音和非话语声音信号是事故调查的重要手段。CVR记录的声音信号受多种因素影响而难以准确识别。目前国外对该信号的应用处于起步阶段，只是通过某一航空器事故案例进行分析，没有形成一套完整、系统的分析方法和技术。国内利用小波技术仅分析了部分开关声的声学特征，这影响着航空器事故原因的深入调查。本研究旨在利用哈希表、美尔倒谱系数及其差分技术，将CVR记录的话语声音信号和非话语声音信号进行音源识别和说话人（机组成员和空中交通管制员）识别，开展CVR记录识别方法的基础研究，提出针对多种不同类型信号的识别方法，从而更加有效、准确确定典型航空器事故发生的情境和发生原因。

中文关键词： 航空器事故调查；驾驶舱话音记录器；音频识别；说话人识别；

英文摘要： The background sound and voice information recorded in aircraft Cockpit Voice Recorders (CVRs) have long been considered as a crucial source of information in addition to flight data in aircraft accident investigations since in most cases they are the latent signal transducers and the only available source of human performance information. Learning from various experiences of the aircraft accident investigation communities of other ICAO members, such as NTSB of United States, MAK of Russia, ATSB of Australia and Taiwan of China, CAAC has been focusing its effort on in depth analysis of different kinds of CVR information. In the aspect of speech information, our research includes development of discourse analysis, speech analysis, and conversation analysis methods based on Chinese pilots'' speech characteristics to evaluate the pilot''s stress level, workload, fatigue, and possible alcohol or drug intoxication. In the aspect of non-speech information, we have applied wavelet transform (WT) and Chirp z-transform (CZT) algorithms to get more accurate spectrum characteristics. Different sound sources from the output of a given aircraft''s CVR can be successfully separated and identified using Adaptive Noise Cancellation (ANC), Blind Signal Separation (BSS), and Back Propagation Neural Network techniques. Knowledge a

英文关键词： Aircraft accident investigation；Cockpit Voice recorder；Sound Identification；Speaker Recognition；