【作者】 郭威；

【导师】 谭云福；

【作者基本信息】 燕山大学 ， 计算机应用技术， 2010， 硕士

【摘要】 目前的语音识别系统普遍采用PC或者服务器的形式作为系统的工作平台,这种方式不可避免地存在体积大、功耗高、不便于携带、实用性低等问题。并且通常的语音识别系统由于噪声、混响等实际情况而导致语音增强处理的过程过于复杂,无法在嵌入式系统中顺利的使用。针对以上问题,本文在总结传统语音增强技术的基础上,展开了对嵌入式语音识别系统的研究,并就课题中所涉及到的相关理论和关键技术进行了深入的探讨,主要包括以下几个方面的工作。首先,介绍了嵌入式系统和语音识别中语音信号增强技术的发展和研究现状,指出了目前语音增强技术存在的问题,论述了课题的主要研究内容;并在介绍几种常用的阵列麦克风拓扑结构设计方案的基础上,全面的分析了各种阵列麦克风语音增强方案的性能指标。其次,研究了一种高效实时的在混响环境下带干扰噪声的语音信号增强方案。该方案以阵列麦克风为前端语音拾取设备,对每个麦克风之间采样得到的语音信号进行多径角度分集接收处理,通过分析语音信号之间的相位关系,多波束形成,对相干信号延时处理并加权合并提高信噪比以实现对采集得到的语音信号的增强处理,并通过调整权值矩阵滤除非语音频段信号和噪声,进而进一步降低可能引入的噪声污染。再次,对系统的硬件平台进行了详细的设计,介绍了嵌入式操作系统的特点及其移植的相关知识;在基于S3C2440的硬件平台上,详细的阐述了系统引导程序BootLoader的编写及Windows CE 6.0的移植过程;并介绍了系统软件的总体设计和关键的语音增强算法的详细研究过程。最后,对系统进行了大量的综合仿真试验,总结系统的各方面能力并分析存在的问题,为进一步的研究提供了方向和宝贵的经验。更多还原

【Abstract】 The current speech recognition system is widely adopted PC or server form as the system work platform. This way inevitably have a lot of problem such as large in size, high power consumption, not easy to carry and low practicality. And the general speech recognition system is usually due to noise, reverberation and other actual conditions lead to voice pretreatment process is too complex to be successfully used in the embedded systems. Against the above problems, this paper summarizes the traditional speech enhancement technology foundation, launched the study of embedded speech recognition system, and conduct a series research of the theoretical and technical which involved in the subject, mainly including the following work.First, this paper introduces the development and research of embedded systems and speech enhancement technology in speech recognition, point out the current problems in speech enhancement technology, and discusses the main aspect of this research topic; Then on the basis of introduce several kinds of array microphone array element design, this paper comprehensive analyse the performance indicators of various microphone array speech enhancement program.Secondly, this paper researched one kind of highly effective real-time program under reverberation environment the belt interference noise voice signal enhancement. This method took the array microphone as the front end speech collection equipment, the voice signal which the sampling obtained to each microphone between carries on multi-diameter angle diversity reception processing, through analyzing the phase relation between the voice signal, the multi-beam formation, delay processing of coherent signals and weights combined enhances the signal-to-noise ratio to realize voice signal enhancement processing which obtained to gathering, and through the adjustment weight matrix filters, only if the speech frequency band signal and the noise, then further reduce the noise pollution which possibly introduces.Again, this paper designed the hardware platform for system in detail, and introduced the embedded operating system features and transplant-related knowledge. Based on S3C2440 hardware platform, this paper described in detail the system boot process which is BootLoader and the transplant process of Windows CE 6.0. Subsequently, this paper introduces the overall design of the system software and key detailed study process of speech enhancement algorithm.Finally, a lot of integrated simulation experiment have been done on the system. Then, this paper summarizes the system of various aspects ability and analyses the existent problems, and provides direction and valuable experience for further study.更多还原