【作者】 李秀珍；

【导师】 仲元昌；

【作者基本信息】 重庆大学 ， 电路与系统， 2010， 硕士

【摘要】 随着科技的发展,人们生活水平的提高,出现了各种各样的家用电器,丰富了人们文化生活、减轻了人们生活负担。如何更有效地管理这些家电,使其更好地为我们的生活服务已成为研究的热点。针对此问题,本课题提出把语音识别技术应用到家电控制中,构建具有语音识别功能的家电集中控制系统,从而实现对家电的集中管理。本文深入研究了语音识别的预处理、特征参数提取、基于连续隐马尔可夫模型(简称CHMM)的训练、识别算法等基本理论。原始语音信号预处理包括:预滤波、预加重、短时加窗、端点检测等;特征参数提取是对预处理后的语音信号提取Mel频率倒谱系数(简称MFCC);训练、识别算法则利用CHMM进行声学建模,建立了基于CHMM的孤立词语音识别算法。研究表明,基于CHMM的语音识别算法在环境噪声干扰的情况下,识别精度显著下降。针对此算法缺陷,从信号空间、特征空间、模型空间三个层次进行语音补偿,构建了一种新的语音识别算法。该算法有效结合了维纳滤波、直方图均衡、向量泰勒级数这三种算法的优点,具有较好的鲁棒性,本文简称该算法为“混合鲁棒语音识别算法”。利用混合鲁棒语音识别算法,采用TI公司的TMS320VC5402为核心芯片,外扩存储电路、语音信号采集电路、LCD显示电路和无线通信模块等;选用电视、DVD、电冰箱、空调、洗衣机、电灯等家电作为控制对象,构建了基于语音识别的家电集中控制平台,最终实现家电的语音控制。系统测试结果表明,在室内噪声环境下,采用混合鲁棒语音识别算法的家电集中控制系统的识别率为98.00%,比没有考虑噪声干扰的基于CHMM的语音识别算法的识别率有显著提高,达到了本课题的预期目的。更多还原

【Abstract】 With the development of technology, various appliances come in our houses, which are enriching people’s lives and reducing the burden of people’s lives. How to manage these appliances more effectively so that they can give us better services for our lives has become a hot issue. For this problem, the paper proposes the appliances can be controlled using the technology of speech recognition, and constructs Appliance Control System.This paper thoroughly studied basic principles of speech recognition, which contins Pre-processing, Feature extraction, Training, Recognition. In the process of Pre-processing, the original input speech signal was followed by the treatments of Pre-filtering, Pre-emphasis, Windowing, Endpoint detection and so on; In order to reduce the redundancy of information, speech signal must be followed by feature extraction, and Mel frequency cepstral coefficients (MFCC) was chosed as the feature parameters; Continuous Density Hidden Markov Models (CHMM) was chosen as the acoustic model of the acoustic unit, and the CHMM-based isolated word speech recognition algorithm has been developed.Studies show that the recognition accuracy of CHMM-based speech recognition algorithm decreased significantly, in noisy environment. To this defect, the paper proposed a new speech recognition algorithm, which was compensated in three levels of signal space, feature space, model space. The algorithm combines the advantages of Wiener Filtering, Histogram Equalization, Vector Taylor Series, and has better robustness, so it is called "the hybrid robust speech recognition algorithm".Taking TMS320VC5402 as the core chip,the writer constructed memory circuits, speech signal acquisition circuits, LCD display circuits and wireless transceiver circuits of Appliance Control System, which realized management of TV, DVD, refrigerators, air conditioners, washing machines, lights and other appliances.The results show that: in interior noise environment, the recognition rate of the appliance control system is 98.00% by the hybrid robust speech recognition algorithm, which reached the purpose of this subject.更多还原