论文部分内容阅读
在传统的基音提取方法中,错误的产生可以分为两类:1)来自分析方法的外在因素,例如由于分析窗的形状、宽度和位置的不适当而产生的不理想的信号表示等等;2)语言信号本身存在的内在因素,例如存在强的谐波分量或次谐波分量等等。实际上,在大多数传统系统中采用的固定帧长和帧的移位对它们的大部分严重基音误差是有影响的。在本文中,我们综合使用活动的波形分析、可数分析窗和可变帧率来解决第一类错误;另外用几个新方法来对付第二种情况产生的误差。利用男、女发音员的语言材料进行试验,结果证实了我们所提出的方法的正确性。
In traditional pitch extraction methods, errors can be classified into two categories: 1) External factors from the analysis method, such as undesirable signal representations due to inappropriate shape and width of the analysis window, etc. ; 2) intrinsic factors that exist in the language signal itself, such as the existence of strong harmonic components or sub-harmonic components and so on. In fact, the fixed frame lengths and frame shifts employed in most conventional systems have an impact on most of their severe pitch errors. In this article, we use a combination of active waveform analysis, countable analysis of windows and variable frame rates to solve the first type of error; in addition to using a few new methods to deal with the second case of error. Using the language materials of male and female speakers, the results confirm the correctness of our proposed method.