听觉掩蔽效应以及语音增强方法研究
文献类型：学位论文
| 作者 | 卜凡亮 |
| 学位类别 | 博士 |
| 答辩日期 | 2002 |
| 授予单位 | 中国科学院声学研究所物理学 |
| 授予地点 | 中国科学院声学研究所物理学 |
| 关键词 | 听觉掩蔽 听觉感知 语音增强 |
| 其他题名 | Hearing masking effect and speech enhancement |
| 中文摘要 | 本工作主要在语音信号的听觉掩蔽效应、带噪语音信号中的噪声参数估计以及基于听觉掩蔽效应特性对带噪语音信号的增强等方面进行了研究。在基于听觉掩蔽的语音增强中，只利用带噪语音对纯音指数的估算难度很大，尤其是当带噪语音的信噪比很低时更是如此。通过听音实验，得到一组保守的听觉掩蔽偏移量，用该组偏移量替代Johnston模型中的听觉掩蔽偏移量可得到一种简化的听觉掩蔽模型。在各种语音增强方法中，背景噪声参数的估计起着重要作用。噪声参数估计的准确与否直接会影响语音增强效果。基于语音信号短时平稳性和简单的功率谱相减法提出一种带噪语音中背景噪声参数的估计方法。基于频谱筛选的语音增强方法，研究了短时幅度谱估计方法及重建效果。根据掩蔽阑和噪声之间的相对关系对带噪语音谱分量有选择地进行处理。如果噪声谱处于掩蔽阑之下则噪声不可闻，一般不需要进行处理；如果噪声谱在掩蔽阑之上则噪声可闻，此时再用某种语音增强方法对它进行处理。引入听觉感知的目的在于最大限度地消除带噪语音中的可闻噪声。但是，已有的基于听觉感知的语音增强方法存在过于强调噪声抑制而忽视由此带来的语音失真的缺点。利用增强语音与纯挣语音的总误差对语音增强效果的分析方法，当背景噪声比较强时要想完全消除可闻噪声是不可能的，提出一种最小感知失真意义下的语音增强准则一气OC语音增强准则。在ADC语音增强准则的基础上，导出语音增强时短时幅度谱的增益函数一ADC增益函数。将ADC增益函数、简单功率谱相减法及简化的听觉感知模型相结合，提出一种基于听觉感知的语音增强方法'一ADc语音增强法。利用听觉系统的掩蔽特性，提出了一种优化的语音增强方法，为了在减少语音失真和加强噪声抑制之间取得良好的折衷，分两种情形对语音信号的幅度谱进行估计。 |
| 英文摘要 | This research work includes three instinct parts: (1) enhancement of speech signal based on characteristic of auditory masking effect; (2) parameter estimation of noise in speech signal;(3) auditory masking effect in speech signal. A new simplified auditory masking model is set up. It is difficult to estimate tone index using speech signal with noise, especially when SNR is very low. A group of conservative offset variables of auditory masking are gotten by tone hearing experiment and a new simplified auditory masking model is set up by replacing offset variables of auditory masking in Johnston model with that group of conservative offset variables. It plays very important role to estimate the parameters of background noise in various speech enhancement methods. A new method is proposed to estimate parameters of background noise in speech signal based on characteristic of short-time smooth of speech signal and simple power spectrum subtraction method. The estimation method of short-time amplitude spectrum and its reconstruction effect is studied based on frequency selection speech enhancement methods. The spectrum of speech signal with noise is processed selectively according to the relationship between noise and masking threshold. Generally specking, when noise spectrum is under the masking threshold, it cannot be heard and must not be treated. When noise spectrum is above the masking threshold, it can be heard and must be treated by some speech enhancement methods. However, most of speech enhancement methods overemphasize noise suppression and ignore speech distortion .A new speech enhancement rule (ADC speech enhancement rule) is proposed on the basis of minimum perception distortion and ADC gain function is introduced. Combining ADC gain function, simple power spectrum subtraction and simplified hearing perception model, a new speech enhancement method (ADC speech enhancement method) based on hearing perception is proposed. An optimal speech enhancement method is proposed by masking characteristic of hearing system. In order to achieve the tradeoff between suppressing speech distortion and enhancing noise reduction, amplitude spectrum of speech signal is estimated in two situations. |
| 语种 | 中文 |
| 公开日期 | 2011-05-07 |
| 页码 | 69 |
| 源URL | [http://159.226.59.140/handle/311008/1022]  |
| 专题 | 声学研究所_声学所博硕士学位论文_1981-2009博硕士学位论文 |
| 推荐引用方式 GB/T 7714 | 卜凡亮. 听觉掩蔽效应以及语音增强方法研究[D]. 中国科学院声学研究所物理学. 中国科学院声学研究所物理学. 2002. |
入库方式： OAI收割
来源：声学研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。