一种新型基频变窗音频信号分析/合成系统

A new audio analysis/synthesis system of adapted window controlled by estimated fundamental frequency

音频信号短时谱的基频随时间会发生变化,因此其谐波成分之间的间隔也会发生变化,在时域上信号随时间会发生或快或慢的变化,这导致短时谱分析所要求的时域和频域分辨率随时间是变化的。传统的固定分析窗由于其时频分辨率固定,无法同时满足上述要求,因而对短时分析造成偏差。本文基于正弦加噪声模型提出了一个分析窗宽受基频控制的自适应新型音频信号分析/合成系统方案,有效地提高了对信号实时分析的精度。并在此基础上,进一步对分析窗的使用、正弦成分的确定和追踪以及噪声成分的分离提出了新的算法和理论依据。本系统对实现音频信号的人为改造提供一套灵活高效的系统框架基础。

The fundamental frequency of short time spectral always varies with the real times, which results in the span between the harmonics varies, while the time signal shows fast or slow variations. The changes of resolution of time or frequency are required in the situation illustrated above. The conventional fixed windows can not satisfy the requirement of resolution both in time and frequency, as leads to the mistake in the analysis of short time. This paper proposes a new adapted audio analysis/synthesis system based on deterministic plus stochastic model in which the size of analysis window is controlled by estimated fundamental frequency in order to improve the effect of analysis for partials. New algorithms and theoretical methods are taken in the design of analysis window, partials determination and tracking and the department of residuals. This scheme offers a robust alternative as the flexible efficient fundamental frame for the music modifying.