智能门禁系统声纹识别中端点检测算法研究

发布时间：2018-07-16 18:05

【摘要】：现代智能化家庭门禁系统声纹识别中端点检测的精度高低直接影响着声纹识别系统的性能。声纹识别系统中高效的语音信号端点检测研究是一个非常重要的研究课题。研究表明,一半以上的语音识别错误是由语音信号的端点检测错误导致的。声纹识别系统中的高效的语音信号端点检测,及端点检测过程中的实时性对声纹识别这项技术能更好的服务于智能化的生产和生活起着关键的作用。但是实际环境中难免会有各种噪声和干扰,使得在对语音信号进行分析和处理时必须考虑噪声的影响。在信噪比较低的场合,语音信号的检测效果严重下降,为了提高声纹识别的准确性,低信噪比下语音信号的端点检测非常有必要。本文首先论述了声纹识别系统中的端点检测原理,概述了语音信号的相关理论知识,综述了现有常见的语音端点检测算法,并对其检测效果进行了仿真验证。仿真与分析证明上述算法在高信噪比条件下,能很好的实现语音信号的端点检测,但是在信噪比较低或是复杂噪声环境下,检测性能会显著下降。针对上述问题,本文在研究已有算法的基础上提出一种基于对数能量加权分带谱熵(Logarithmic Energy weighted Band-partitioning Spectral Entropy feature,简称LESEf)的语音信号端点检测算法,该算法结合了对数能量以及分带谱熵在语音信号端点检测中的优势,改善了因为幅值问题导致语音与噪声误判的情况,同时克服了单个频率点上噪声对语音的影响,提高了端点检测的准确性,具有较高的鲁棒性。实验和分析结果表明,本文算法能实现低信噪比下信号端点的准确检测,并且检测效果要好于现有的基于对数能量(Logarithmic Energy,简称LE)的语音信号端点检测算法。
[Abstract]:The accuracy of the endpoint detection in the modern intelligent home entrance guard system has a direct impact on the performance of the acoustic stripe recognition system. It is a very important research topic to study the effective endpoint detection of voice signal in voiceprint recognition system. Research shows that more than half of the speech recognition errors are caused by endpoint detection errors of speech signals. The efficient endpoint detection of voice signal and the real time in the detection process of voiceprint recognition system play a key role in the intelligent production and life of voiceprint recognition technology. However, it is inevitable that there will be a variety of noise and interference in the actual environment, which makes it necessary to consider the influence of noise in the analysis and processing of speech signals. In order to improve the accuracy of voiceprint recognition, it is necessary to detect the endpoint of speech signal at low SNR. This paper first discusses the principle of endpoint detection in voiceprint recognition system, summarizes the relevant theoretical knowledge of speech signal, summarizes the existing common speech endpoint detection algorithms, and verifies the detection effect by simulation. Simulation and analysis show that the proposed algorithm can achieve the endpoint detection of speech signal well under the condition of high signal-to-noise ratio (SNR), but the detection performance will be significantly decreased under the condition of low SNR or complex noise. In order to solve the above problems, this paper proposes a speech signal endpoint detection algorithm based on logarithmic Energy weighted Band-partitioning Spectral feature Entropy (LESEf). The algorithm combines the advantages of logarithmic energy and band spectrum entropy in the endpoint detection of speech signals, improves the misjudgment of speech and noise caused by amplitude problems, and overcomes the influence of noise on speech at a single frequency point. The accuracy of endpoint detection is improved and the robustness is high. The experimental and analytical results show that the proposed algorithm can accurately detect the signal endpoints at low signal-to-noise ratio and is better than the existing logarithmic energy (le) based speech signal endpoint detection algorithms.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN912.3

【相似文献】