带噪语音信号端点检测算法的研究及实现

发布时间：2018-04-20 18:50

本文选题：谱熵 + 谱减法　；参考：《河南理工大学》2014年硕士论文

【摘要】：一个完整的语音识别系统,其最终实现及使用效果不仅限于识别算法,许多相关因素也会影响语音识别系统的成功。语音识别的对象是语音信号,作为识别中的第一个重要环节,端点检测就是要对复杂的环境中的语音信号和非语音信号进行区分,确定有用信号的开始和结束。语音信号通常都存在一定的背景音,在实际环境中,其不可避免地受到各种噪声的影响,语音端点检测是语音识别的第一个关键技术,只有提高语音信号端点检测的精确度,才可以进一步提高语音识别的准确率。本文在已有端点检测算法的基础上,重点研究了以下两点:其一,在MATLAB仿真环境中对各种语音端点检测算法进行了研究,在基于谱熵端点检测算法中,提出了改进的的归一化加权因子。由谱熵的优势和特点,证明了算法的有效性和稳健性。为下文运用谱熵来进行端点检测提供了一种很适合的算法。其二,对基于语音增强的端点检测算法中,重点研究了LMS算法和谱减法。在基本谱减法的基础上,提出了改进谱减法,在解决噪声复杂问题上,引入了自适应噪声平滑。结合改进谱减法和自适应噪声平滑,对含噪语音信号进行端点检测。仿真结果表明,改进算法更能够适应复杂噪声环境,提高了语音信号端点检测的精确度。
[Abstract]:A complete speech recognition system, its final implementation and use effect is not only limited to the recognition algorithm, but also many related factors will affect the success of the speech recognition system. The object of speech recognition is speech signal. As the first important link in recognition, endpoint detection is to distinguish speech signal from non-speech signal in complex environment and to determine the beginning and end of useful signal. Speech signal usually has a certain background sound. In the actual environment, it is inevitably affected by various kinds of noise. Speech endpoint detection is the first key technology of speech recognition, only to improve the accuracy of speech signal endpoint detection. The accuracy of speech recognition can be further improved. Based on the existing endpoint detection algorithms, this paper focuses on the following two points: first, in the MATLAB simulation environment, a variety of speech endpoint detection algorithms are studied, in the spectral entropy based endpoint detection algorithm, An improved normalized weighting factor is proposed. The effectiveness and robustness of the algorithm are proved by the advantages and characteristics of spectral entropy. It provides a suitable algorithm for endpoint detection using spectral entropy. Secondly, the LMS algorithm and spectral subtraction are studied in the endpoint detection algorithm based on speech enhancement. Based on the basic spectral subtraction, an improved spectral subtraction is proposed, and adaptive noise smoothing is introduced to solve the complex noise problem. Combined with improved spectral subtraction and adaptive noise smoothing, endpoint detection of noisy speech signal is carried out. Simulation results show that the improved algorithm can adapt to the complex noise environment and improve the accuracy of speech signal endpoint detection.
【学位授予单位】：河南理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN912.34

【相似文献】