基于自适应频率尺度变换的骨导鼾声识别关键技术研究

项目名称： 基于自适应频率尺度变换的骨导鼾声识别关键技术研究

项目编号： No.61471259

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 无线电电子学、电信技术

项目作者： 魏建国

作者单位： 天津大学

项目金额： 85万元

中文摘要： 打鼾是睡眠过程中呼吸削弱的征兆,20%的人都会打鼾, 其中15%的打鼾者患有睡眠呼吸暂停综合症,其死亡率高达40%，我国大约有 3750 万人的健康受到该病症威胁。本课题对利用骨导音采集的鼾声进行自动识别中的关键技术进行研究。包括对于骨导鼾声采集位置的研究；建立大规模骨导鼾声音频及同步体态数据库。在鼾声特征参数提取方面，根据鼾声、呼吸信号、语音及咳嗽等信号分布计算F-ratio，按照每个频带对鼾声识别的贡献率的大小重新安排滤波器的分布，提出鼾声自适应频率尺度变换新方法，从而更好的获取鼾声音频信号特征。利用受限波尔兹曼机对鼾声特征向量进行降维研究，将原始数据特征向量中各个元素间的相关性体现到高层特征表示中，并降低特征参数维数。通过同步采集的睡姿及鼾声信号，定量分析两者间相关性。利用睡眠呼吸行为事件的上下文关系建立鼾声行为统计模型，并利用隐马尔可夫模型方法建立鼾识别验证系统。

中文关键词： 鼾声检测；骨导音；自适应频率尺度变换

英文摘要： Snoring is the sign of weak breathing during sleep. Nearly 20% of people snore while sleeping. Snoring has become a big problem that endangers people's health. About 15% of population who snore suffers Sleep Apnea Syndrome, which causes a mortality rate of 40%. There are 37.5 million people affected by this health threat in China. This study will focuses on developing the key techniques of detection of snore based on bone conduction sound. Detailed topics include: 1) Building a snore database by using a bone conducted snore detection device developed by our team. 2) Measuring the dependences between frequency components and the characteristics of bone conducted snore sound, so as that we adopt an adaptive frequency filter to extract more discriminative feature of BC snore signal. 3) Restricted Boltzmann Machines is used to reduce the dimensions of feature vectors of snoring signal. Also this method is expected to improve the protection of privacy information among snore signal. In addition, snoring statistic model is established by the context of sleep action events. 4) Meanwhile, snore sound recognition system will also be established by using HMM based method.

英文关键词： Snore recognition;Bone conducted sound;Adaptive frequence coefficients