| 跟着形式辨认、人工智能等范畴的赓续成长，法语论文网站，人们对语音旌旗灯号处置后果和语音质量的请求赓续进步，然则，在实际技巧应用中，语音旌旗灯号弗成防止地遭到各类噪声的搅扰，噪声的存在严重作用语音旌旗灯号处置体系的机能。语音加强就是专门研究从带噪语音旌旗灯号中尽量无掉真恢回复复兴始语音的技巧。若何简略且高效地清除噪声的作用一向是语音加强技巧的一年夜困难，谱衰减语音加强算法是一种机能优良并且易于完成的办法，在这一框架下有许多的语音加强办法都获得了不错的后果。然则，谱衰减算法中先验信噪比的精确估量成绩和语音掉真成绩一向未能很好处理。本论文针对这些成绩重要展开了以下研究(1)剖析了传统谱减法语音加强后发生“音乐噪声”的缘由，在此基本上提出一种基于先验信噪比的谱减法。起首，采取先验信噪比估量算法获得先验信噪比，接着，依据先验信噪比和后验信噪比之间的关系，将传统谱减法增益函数中的后验信噪比用先验信噪比替换获得基于先验信噪比的增益函数。运用该办法对带噪语音处置能有用克制“音乐噪声”，法语毕业论文，进步语音可懂度。(2)提出运用MMSE来估量先验信噪比停止语音加强。在树立语音和噪声模子的基本上，运用最小均方误差准则直接从带噪旌旗灯号中估量先验信噪比。经由过程仿真试验比较，证实该办法能有用战胜DD(decision一directed)先验信噪比估量办法存在估量滞后一帧和非因果先验信噪比估量办法不具及时性的缺陷，说明了该办法的有用性。(3)针对语音加强时谐波成份丧失招致语音掉真的成绩，提出自顺应谐波再生语音加强办法。经由过程对加强语音停止非线性变换获得谐波恢覆信号，联合信噪比相干的自顺应因子，获得一个由加强语音协调波旌旗灯号调剂的谐波再生加强语音。它有用地战胜了采取固定调剂因子的传统谐波再生办法对信噪比拟低区域恢复才能缺乏的缺点。经由过程不雅察语谱图和停止可懂度测试等，证实了该办法能有用地恢复加强语音中丧失的谐波成份，削减语音掉真。 Abstract: Follow the form of identification, artificial intelligence and other areas develops ceaselessly, but people of progress, request to speech signal processing and speech quality consequences, in practical skills, the speech signal is inevitable by all kinds of noise interference, has seriously affected the speech signal processing system noise function. Speech enhancement is a special seminar from the noisy speech signal without distortion to restore the Renaissance speech skills. It has been a great difficulty to remove noise from the speech enhancement technique. The speech enhancement algorithm is a kind of method which has good performance and is easy to be done. However, the accurate estimation of the prior SNR of spectral attenuation algorithm has been unable to deal with the real results of speech. In this thesis, the following research (1) has been carried out to analyze the reason of "music noise" after the traditional spectral subtraction. First, a priori SNR is adopted to obtain a priori SNR, then the gain function based on a priori SNR is obtained by replacing the prior SNR with the prior SNR. Application of the approach to the disposal of noisy speech can be useful to control the music noise, the progress of speech intelligibility. (2) the application of MMSE to measure a priori SNR is proposed to stop the speech enhancement. In a speech and noise model basically, the application of the principle of minimum mean square error directly from the noisy signal in estimating a priori snr. Through simulation experiments, it is proved that the proposed method can overcome the defect of the prior SNR estimation of DD (directed decision), which is not timely, and the usefulness of this method is proved. (3) for the enhancement of the speech enhancement, the harmonic components are lost to the speech and the speech enhancement method is proposed. Through the process of strengthening the speech nonlinear transform of harmonic signal restore, self adaptation factor than coherent combined signal-to-noise, obtain a speech by strengthening coordination wave signal harmonic regeneration speech strengthen adjustment. It is useful to overcome the traditional harmonic regeneration method to take the fixed swap factor for the low SNR of the low recovery of the region can be a lack of shortcomings. Through the process of observing the spectrogram and stop the intelligibility test, confirmed that the method can effectively enhance the recovery of the loss of the harmonic components of speech, speech really cut off. 目录: 摘要 4-5 Abstract 5-6 第1章 绪论 9-18 1.1 语音增强技术的背景和意义 9-10 1.2 语音增强技术的探讨近况与存在的问题 10-16 1.2.1 语音统计模型 10-11 1.2.2 语音增强算法 11-13 1.2.3 噪声估计 13-14 1.2.4 语音评价 14-16 1.3 论文主要工作 16 1.4 论文的主要探讨内容与章节安排 16-18 第2章 谱衰减语音增强算法 18-27 2.1 原理 18-19 2.2 谱衰减语音增强算法 19-22 2.2.1 谱减法 19-20 2.2.2 维纳滤波 20-21 2.2.3 最小均方误差估计(MMSE) 21 2.2.4 其它谱衰减法 21-22 2.3 作用谱衰减语音增强算法效果因素略论 22 2.4 基于先验信噪比估计的谱减法 22-23 2.5 仿真实验与略论 23-26 2.5.1 三种谱衰减语音增强措施增强效果对比 23-24 2.5.2 传统谱减法与基于先验信噪比估计的谱减法增强效果对比 24-26 2.6 本章小结 26-27 第3章 基于MMSE 先验信噪比估计的语音增强措施 27-36 3.1 DD 先验信噪比估计 27-28 3.2 非因果先验信噪比估计 28-30 3.3 MMSE 先验信噪比估计 30-31 3.4 仿真实验与结果略论 31-35 3.4.1 信噪比略论 31-32 3.4.2 失真度略论 32-33 3.4.3 信号时域波形和语谱图略论 33-35 3.5 本章小结 35-36 第4章 自适应谐波再生语音增强措施探讨 36-48 4.1 谐波再生语音增强 36-38 4.2 自适应谐波再生语音增强 38-41 4.4 仿真实验与结果略论 41-47 4.4.1 HRNR 与AHRNR 效果比较 41-42 4.4.2 不同增强算法与AHRNR 结合 42-45 4.4.3 不同先验信噪比估计算法与AHRNR 结合 45-46 4.4.4 AHRNR 加MMSE 先验信噪比估计措施语音增强实验 46-47 4.5 本章小结 47-48 第5章 总结和展望 48-50 5.1 全文工作总结 48-49 5.2 展望 49-50 参考文献 50-55 致谢 55-56 附录A 个人简历 56-57 附录B 攻读硕士学位期间撰写的论文 57-58 附录C 论文中的用图 58-59 附录D 论文中的用表 59 |
