论文部分内容阅读
非平稳声信号的接续的短时间谱的估计常常以强度调制的频率-时间图(“谱图”)的形式显示。本文研究了在把某些形式的谱图当作数字图象来存储和处理中的一些问题和优点。我们所讨论的特定形式的谱图是从产生时变线谱的声源得到的,这种时变线谱由一根占支配地位的主线和几根弱得多的谐频或准一谐频线所组成。这一类型的谱图通常是由旋转机械产生的。假定在测量这种信号时混有加性宽带噪声,所得到的谱图是由一族相互关联的弯曲的线所构成的,其中那些较弱的线趋向于淹没在噪声之中。在所提出的图象分析方案中,原始的谱估计是作为一个数字谱图来存储的。然后,用线谱增强方法来处理这幅谱图以提取主线,从而形成“族样板”。利用这种样板可以检测和提取弱线。通过把提取出来的各个分量加以重新组合,便得到一幅经过净化的信号低熵频率-时间图象,它比原来的谱图能更好地表示出信号的结构特点,更适合于作自动分析.
Estimates of successive short-term spectra of non-stationary acoustic signals are often displayed as intensity-modulated frequency-time plots (“spectra”). This article examines some of the problems and benefits of storing and processing some forms of spectrograms as digital images. The particular form of spectrum we are discussing is derived from a source that produces a time-varying line spectrum consisting of a dominant dominant line and a few much weaker harmonic or quasi-harmonic lines Composed of. This type of spectrum is usually generated by a rotating machine. Assuming that this signal is mixed with additive broadband noise, the resulting spectrum is made up of a family of interlinked curved lines, of which the weaker lines tend to submerge in the noise. In the proposed image analysis scheme, the original spectral estimate is stored as a digital spectrum. Then, the line spectrum enhancement method to deal with this spectrum to extract the main line, resulting in “family template ”. Use this template to detect and extract lines of weakness. By recombining the extracted components, a clean, low-entropy frequency-time image of the signal is obtained, which shows the structural characteristics of the signal better than the original one and is more suitable for automatic analysis .