基于自适应时频分析的介质声参数反演及语音信号分析
文献类型：学位论文
| 作者 | 张德明 |
| 学位类别 | 博士 |
| 答辩日期 | 2005 |
| 授予单位 | 中国科学院声学研究所 |
| 授予地点 | 中国科学院声学研究所 |
| 关键词 | 自适应最优核时频分析 线性时变滤波 色散 非平稳 |
| 其他题名 | Medium Acoustical Parameters Inversion and Speech Signal Analysis Based on Adaptive Time Frequency Analysis |
| 中文摘要 | 对于难以直接测量的系统来说，可以通过对该系统输入一已知的特定信号，测量并分析其输出信号来获得关于这一系统的若干特性。而输入信号可分为稳态信号和瞬态信号或脉冲，与之相联系的系统输出就称为稳态响应和瞬态响应，很多情况下系统的非平稳瞬态响应更能揭示我们所感兴趣的系统特征。例如在水声领域中，传统的方法是利用垂直阵接收一定距离外的脉冲声信号来反演海底及相关水文的参数，在地声领域也有通过阵列接收的瞬态信号进行地声参数反演的诸多应用。本文提出了基于自适应最优核（AOK）时频分析（TFA）及时变滤波（TVF）的介质声参数反演方法，是完全基于单传感器信号处理的，相对于传统的阵列处理的方法来说，不但具有较高的时频分辨性能，而且不存在通道一致性问题，可以准确给出多模式信号的色散特性，还可以将时频面中非重叠的模式波形提取出来；另外本文还提出了基于时频分析及时变滤波的双通道语音分离算法，与传统盲源分离算法所依赖的语音频谱统计平稳的假设不同，本文的算法则是完全基于语音非平稳特性的。对于另外两种应用较为广泛的时频分析方法，即小波变换〔WT）方法和时频重排（TFRA）方法，本文对它们的时频分析的本质及性能进行了讨论，对其时频分析局限性给出了合理的解释；并和AOK方法的结果进行了对比，可以发现AOK方法在时频聚集性能及抗噪声性能上都有较好的表现。并将其首次运用于浅海简正波的色散表征、模式提取及以此为基础的声参数反演，以及语音共振峰提取，双通道的语音分离，并且取得了较为可靠的结果。 |
| 英文摘要 | In order to measure the parameters of an unknown system, analysis of the output of the system corresponding to a certain input signal, is a common approach. The input signal can be steady or transient, so the response of correspondence can be called steady or transient response. In many cases, the non-stationary transient response can disclosure more details about the system properties concerned. In underwater acoustics, the traditional approach of medium acoustical parameters inversion is fulfilled by employing a vertical sensor array, which receives the impulsive signal from a long distance. In geo-acoustics, there are also many applications on acoustical parameters inversion by means of analysis of the non-stationary signal received by a sensor array. A new approach for medium acoustical parameters inversion is proposed, which is based on adaptive optimal kernel (AOK) time frequency analysis (TFA) and linear time variant (LTV) filtering. The signal analysis and processing is totally based on a single sensor, it has no channel consistency problem, other than a sensor array encounters in traditional approach. The proposed method has high time frequency resolution, can provide precise dispersion curves of the wave modes in shallow water or in solid surface. And the waveform of a single mode not interfered with others, can be extracted by LTV filtering. A new co-channel speech separation algorithm is also proposed, which based on AOK method and LTV filtering. Another two methods of acoustical parameters inversion also based on single sensor, wavelet method and time frequency rearrangement (TFRA) method, are also discussed, especially on their time frequency nature, which results in their poor performance in time frequency analysis. For the same non-stationary signal, the time frequency analysis resolution performance of wavelet method, TFRA method and AOK method are compared. The results show the high performance of AOK as to its high time frequency-resolution and robust against the noise. Two practical experiments and the results are provided, the first is the acoustical parameters inversion of shallow water bottom, second is co-channel speech separation. The processing results show the efficiency of the proposed approach. |
| 语种 | 中文 |
| 公开日期 | 2011-05-07 |
| 页码 | 85 |
| 源URL | [http://159.226.59.140/handle/311008/886]  |
| 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 | 张德明. 基于自适应时频分析的介质声参数反演及语音信号分析[D]. 中国科学院声学研究所. 中国科学院声学研究所. 2005. |
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。