低信噪比条件下的语音信号检测

发布时间：2018-06-13 14:22

  本文选题：信号处理 + 自适应学习　； 参考：《杭州电子科技大学》2017年硕士论文

【摘要】：传统语音信号处理直接利用语音信号的时域参数并加以频谱参数辅助判决以达到改善语音质量的目的,但是针对实际情况中噪声强且非平稳的特性,其较弱的鲁棒性导致系统性能受信噪比的影响较为敏感。本文首先从低信噪比条件下语音信号检测的应用领域、实用价值以及研究现状等角度对国内外相关文献进行梳理,剖析了强噪声环境下语音信号模型的复杂性,总结了非平稳噪声环境中语音信号检测的关键问题。其次,研究改进了基于短时信噪比的自适应阈值和自适应判决语音端点检测算法,根据自适应短时能量,并加以短时过零率和自适应判决校验,得到最终的端点检测结果。然后,针对语音信号在低信噪比条件下结构的复杂性,基于自适应学习和基本谱减法研究改进了一种基于子带谱熵的语音增强算法,该算法将带噪语音分成若干个子带分别进行自适应加权,并计算子带谱熵值以用来估计噪声谱能量。最后,实验结果表明,自适应阈值端点检测算法在不同信噪比的平稳噪声和非平稳噪声中均能有效检测出语音中的有话段和无话段之间的端点,且算法准确性和鲁棒性明显优于传统的端点检测算法,另外,自适应子带谱熵语音增强算法在不同真实混合噪声源分别影响下,同样能达到较优的性能,且对语音信号质量提高效果显著。
[Abstract]:The traditional speech signal processing directly utilizes the time domain parameters of the speech signal and adjusts the spectrum parameters to improve the speech quality, but in view of the characteristics of strong noise and non-stationary in the actual situation, Because of its weak robustness, the system performance is sensitive to the influence of signal-to-noise ratio (SNR). In this paper, the application field, practical value and research status of speech signal detection under low signal-to-noise ratio (SNR) are firstly analyzed, and the complexity of speech signal model in strong noise environment is analyzed. The key problems of speech signal detection in non-stationary noise environment are summarized. Secondly, the adaptive threshold and adaptive decision speech endpoint detection algorithm based on short time signal-to-noise ratio (SNR) are improved. According to the adaptive short time energy and the short time zero crossing rate and adaptive decision check, the final endpoint detection results are obtained. Then, aiming at the complexity of speech signal structure under low SNR, a speech enhancement algorithm based on sub-band spectral entropy is proposed based on adaptive learning and basic spectral subtraction. The algorithm divides the noisy speech into several sub-bands for adaptive weighting and calculates the entropy value of the sub-band spectrum to estimate the noise spectral energy. Finally, the experimental results show that the adaptive threshold endpoint detection algorithm can effectively detect the endpoint between the speech segment and the non-speech segment in different SNR stationary noise and non-stationary noise. The accuracy and robustness of the algorithm are obviously superior to those of the traditional endpoint detection algorithm. In addition, the adaptive sub-band spectrum entropy speech enhancement algorithm can also achieve better performance under the influence of different real mixed noise sources. And the effect of improving the quality of speech signal is remarkable.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.3
，

本文编号：2014312