论文部分内容阅读
作为混沌信号描述的技术,分形维数能够对信号的规律与周期进行度量。语音比白噪声的精细度偏小与规则性较高,正是这一个方面的原因,所以分形端点检测能够充分区分白噪声与语音。然而,针对某些特定的噪声来说,其精细度比语音小,而规则性较高。所以,针对复杂多变的噪声条件,该方法不能获得良好的区分结果。为进一步改善该方法的鲁棒性,确保该方法对各种噪声均能够适用,阐明了基于短时频域的分形端点检测算法。这个方法在区分噪声与语音的过程中主要从短时频域能量分布上提取分形维数来进行。通过实验发现,设计的方法的鲁棒性相对较好,一方面对那些没有规律的白噪声适合,另一方面对那些时域规律性与周期性相对较强的噪声具有较好的适用性。
As a technique of chaotic signal description, fractal dimension can measure the law and period of signal. The reason that the speech is smaller and more regular than the white noise is the reason of this aspect. Therefore, the fractal endpoint detection can fully distinguish the white noise from the speech. However, for some specific noise, its fineness is lower than that of the speech, and the rule is higher. Therefore, the method can not obtain good discrimination results under complex and variable noise conditions. In order to further improve the robustness of this method and ensure that this method can be applied to all kinds of noise, the fractal endpoint detection algorithm based on short-time frequency domain is expounded. This method is mainly used to extract the fractal dimension from the energy distribution in short-time frequency domain in the process of distinguishing noise from speech. The experimental results show that the robustness of the proposed method is relatively good, on the one hand it is suitable for those irregular white noises, on the other hand, it has good applicability to the noise with the regularity in the time domain and the relatively strong periodicity.