基于BP网络和多抽样率处理的缺失音频信号恢复方法

Missing audio signal recovery approach based on BP neural network and multirate processing

提出并设计了一种由多个BP网络和均匀DFT滤波器组构成的非平稳序列自适应预测器,用于缺失音频信号的恢复.通过改进样本处理方法和分别设计各个BP网络结构,达到较高的预测精度.采用多抽样率并行处理方式,减少了单个BP网络的数据处理量,提高了预测器恢复缺失的音频信号速度.仿真实验证明效果较好.

A nonstationary sequence predictor for missing audio signal recovery is presented and designed. The predictor consists of a few BP networks and a uniform DFT filter bank. Through the improved method of sample processing and designing every BP network respectively, the predictor reaches a high precision comparatively. The multirate parallel processing reduces the data quantity processed by one BP network and increases the speed of missing audio signal recovery. Stimulation results show that this method is effective and efficient.