语音识别技术在语言教学软件中的应用研究

【作者】 熊飞丽；

【导师】 张玘；

【作者基本信息】 国防科学技术大学 ， 精密仪器及机械， 2002， 硕士

【摘要】 随着计算机技术的发展，计算机辅助教学已成为现代教育技术在教育领域运用的一个重要方面。越来越多的学习软件已经在帮助人们学习语言。计算机丰富的图形、图象、声音处理功能有力促进了人们的语言学习效果。因此，探索有效的语言学习方法，开发具有语音识别／判别能力的教学软件，把语音识别技术与多媒体技术相结合，已成为这一类语言教学的热点。 作者在本论文中，对国内外语音识别技术发展状况做了较全面的总结分析，对语音信号产生模型、线性预测编码方法、求解LPC正则方程的德宾递推算法、语音信号同态处理方法、LPC倒谱特征计算、动态特征匹配等语音识别的关键环节的技术问题进行了深入的理论分析和仿真研究，用Matlab语言编写了语音信号滤波、分帧、特征计算和匹配软件，并给出了仿真计算结果。实验结果表明，与采用LPC特征相比，采用LPC倒谱特征和动态匹配算法进行短时语音识别，会有较高的识别率；对不同语音信号有特征空间离散度大、易于确定判别门限的特点。特征计算所需要的递推算法也易于在DSP上实现。因此在未来的智能多媒体语言教学系统中，LPC倒谱特征语音识别方法具有较好的应用前景。所做的论文工作，为在DSP上进行语音识别算法开发提供了理论分析与仿真实验依据。更多还原

【Abstract】 With the rapid development and increasing popularization of computer technology,. Computer-Assisted Instruction(CAI) has been widely used in the teaching process. More and more education software become popular in helping people to acquire knowledge,Especially in studying language. The multimedia ability of computer will play an important role in learning featured by vivid voice and picture,as well as the large storage of information. Right now,in the research field of language studying,much more attention has been paid to develop valid voice recognition (abbr. VR)strategy so as to made people study language easier. But it is a pity that language education software has rarely valid voice recognition function. So,in this paper,the theory and algorithm of VR are being developed.In this paper,several key problems in VR process are being discussed both in theory and application,which include pre-processing ,frame decomposing of raw voice signal,characteristic selection and calculation,dynamic mapping of characteristics. Linear prediction model ,model coefficients(LPC) ,as well as cepstrum coefficients are well analyzed both in analysis and calculation aspects . Dynamic mapping algorithm is also illustrated in details. Through the computer simulation to some real short-time voice signal samples using Matlab language. The result shows that the recognition efficiency using cepstrum coefficients mapping is better than what made by LPC mapping . This conclusion is more attractive in the application development of language education system using Digital Signal Processor(DSP).更多还原