| 题名 | 一种改进的MMSE-STSA语音增强方法 |
| 作者 | 张鑫琪 |
| 学位类别 | 博士 |
| 答辩日期 | 2008-05-30 |
| 授予单位 | 中国科学院声学研究所 |
| 授予地点 | 声学研究所 |
| 关键词 | 最小均方误差估计 非平稳噪声 非线性函数 谱增益修正 |
| 其他题名 | An Improved MMSE-STSA Speech Enhancement Method |
| 学位专业 | 信号与信息处理 |
| 中文摘要 | 语音增强的目的是改进语音质量，在消除背景噪声的同时提高语音可懂度。目前一些对非平稳噪声干扰下的语音信号进行增强的方法，可以降低背景噪声，但是有时会引入不舒服的音乐噪声，而且这些方法不能很好的提高语音的可懂度，甚至使其略有下降。基于短时幅度谱估计(STSA，Short-time Spectral Amplitude)的方法作为一种单通道语音增强方法，以其简单有效深受欢迎。 语音增强系统中一个重要组成部分即为噪声功率谱估计。传统的估计方法是在无语音段估计出噪声值，而在有语音段用此值近似代替。为了区分有语音段和无语音段，就需要对带噪语音信号进行语音活动性检测(VAD，voice activity detect)。然而，语音活动性检测的可靠性在弱语音信号以及低输入信噪比的条件下急剧恶化，而且在非平稳噪声环境下严重制约了对背景噪声的追踪性能。 本文通过一个非线性函数，根据带噪语音信号的信噪比对非平稳背景噪声信号进行估计，应用最小均方误差估计方法(MMSE，minimum mean-square error)，利用估计出的噪声功率谱得到相应的谱增益，进而估计出纯净语音信号的短时幅度谱。方法中对谱增益的修正，可以进一步抑制低信噪比时的残留噪声以及降低对带噪语音信号的过抵消。本文用MATLAB实现了整个算法的仿真，并与传统的谱相减法、最小均方误差短时幅度谱估计(MMSE-STSA)方法的增强结果相比较。仿真结果表明，该算法对非平稳噪声的追踪性较好，在抑制背景噪声，减少音乐噪声的同时，提高了语音的可懂度，其计算复杂度也有很大优势，便于该算法在实际中的使用。 |
| 英文摘要 | The main objective of speech enhancement is the improvement in quality and intelligibility of a degraded speech signal while eliminates the background noise, but it often fails to achieve the two objectives at the same time, especially under the non- stationary background noise. At present, some speech enhancement algorithms can reduce the non-stationary background noise, but sometimes they will bring in uncomfortable music noise, and cannot improve the intelligibility, in some cases even decline it. One of the single-channel (one microphone) enhancement methods which based on the short-time spectral amplitude estimation is most welcomed for its simplicity and effective in the practical application. A crucial component of a practical speech enhancement system is the estimation of the noise power spectrum. A common approach is to average the noisy signal over non-speech sections. Speech pause detection is either implemented on a frame-to- -frame basis, and the detection reliability severely deteriorates for non-stationary environment, week speech components and low input SNR. Additionally, the amount of presumable non-speech section in the signal may not be sufficient, which restricts the tracking capability of the noise estimator in case of varying noise spectrum. In this paper, the noise power spectrum is estimated through a nonlinear function in accordance with the estimated SNR. The minimum mean square error method is used to estimate the spectral gain which can calculate the short-time spectral amplitude of the pure speech signal. The spectrum gain modification can further suppress the residual noise for low SNRs and avoid the excessive suppression. The simulation of the whole algorithm and comparison of the results with the traditional spectrum subtraction method and MMSE-STSA method are given in this paper. Under the non-stationary noisy environment, the proposed algorithm can not only get a good performance in enhancement, but also reduce the speech distortion, and its computational complexity has a great advantage to practical use. |
| 语种 | 中文 |
| 公开日期 | 2011-05-07 |
| 页码 | 79 |
| 内容类型 | 学位论文 |
| 源URL | [http://159.226.59.140/handle/311008/428]  |
| 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 | 张鑫琪. 一种改进的MMSE-STSA语音增强方法[D]. 声学研究所. 中国科学院声学研究所. 2008. |
| 个性服务 |
| 查看访问统计 |
| 相关权益政策 |
| 暂无数据 |
| 收藏/分享 |
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。
修改评论