【作者】 蔡虎；

【导师】 邹雪城；

【作者基本信息】 华中科技大学 ， 软件工程， 2009， 硕士

【摘要】 语音识别和嵌入式是当今热点,其产品应用前景十分广阔,将语音识别应用到嵌入式系统中,更是能够发挥两者的优点。本文主要研究嵌入式孤立字语音识别,主要研究设计语音识别算法:动态时间规整算法DTW和隐马尔科夫算法HMM,然后编写了基于linux和arm芯片的音频驱动并对其进行了测试。本文在语音识别算法方面,首先建立了语音识别系统的发声模型,然后对语音进行预处理,即对数字信号预加重、分帧和加窗,接着对语音信号提取特征参数,提取特征参数的方法采用线性预测倒谱系数法和美尔倒谱系数法,再接下来是要对提取出来的特征矢量进行矢量量化,从而压缩数据。最后重点分析设计了动态时间规整DTW(Dynamic Time Warping)识别算法和隐马尔科夫HMM(Hidden Markov Model)识别算法。DTW主要是通过寻找动态时间规整函数来使输入模块与参考模块之间的误差达到最小,从而得出语音识别的相似误差。隐马尔科夫HMM识别算法主要是解决求观察值概率的问题、求最佳状态序列的问题和在什么情况下观察值概率最大的问题。基于以上两个模型基础,本文设计了matlab语音识别代码,得到了期望识别结果。在嵌入式方面,本文采用的操作系统为linux系统,硬件平台使用友善之臂的mini2440开发板,主要使用了开发板上声卡资源和LED资源,以及调试所用的串口。编写了音频驱动/dev/dsp和/dev/mixer,它们分别对应着音频驱动的两个部分,一个是主要用来录音播音的,另一个是用来对声音的大小进行控制。驱动编译成功后,然后加载进系统内核,并通过音频测试程序来测试上述两个设备。在开发应用试验方面,本文是通过语音来控制LED灯,不同的语音信号发出时LED灯就会处于不同的状态,以达到语音控制嵌入式设备的目的。最后提出了语音控制机场地勤服务的应用,并做了一些语音识别技术的研究,为后续研究打下基础。更多还原

【Abstract】 Speech recognition and embedded system are quite important nowadays, there are great values in the field of research and market. For example, they could be applied in the controlled devices and inspected devices with making people coming out of hot and contaminated environment. Furthermore, the speech recognition and embedded system can make great changes in the area of phone in the future because the production with this characteristic will be more humanistic.However, the progress of speech recognition is not easy, it takes a lot of time and energy to research. And it has some different classes in the system of speech recognition. In this paper, it mainly talks about particular word speech recognition, and has used dynamic time warping and hidden markov model.The progress of speech recognition concludes building of speech model, pretreatment of speech signal, detecting endpoint, feature extraction, training and recognition of speech. This paper uses the tools matlab to emulate the progress of the research, so the matlab photes can clear show the progress for speech recognition, and users can understand speech recognition easily.In the other hand, this paper chooses Linux operation to be its embedded operation, and chooses the mini2440 developed board making by armfriendly firm, and its audio chip and led light are used in this paper. This paper also wrote audio drive, and then load into the kernel, and it is well according to the test.In the application of this system, the main work is to control the lights under the speech orders, the lights are in different status in different speech orders, so in future it can control other devices.In recent years, although there are some productions about speech recognition and embedded system, their price are often quite high because of its advanced characteristic. The research of this paper can reduce cost in some degree, and has some values in the practicality.更多还原