基于经验模式分解的音频带宽扩展算法

音频带宽扩展是现代音频编解码器的重要组成模块,它可以大幅降低编码码率。现有方法大多将低频复制到高频,然后利用少量高频参数进行调整得到重建的高频信号。这种复制方法假设高低频信号间具有相关性;但是,当这种相关性变弱时,编码质量急剧下降。为解决该问题,提出一种经验模式分解的方法重建高频信号。首先对高频和低频激励信号进行经验模式分解,然后选择与高频模式分量相关度较大的低频模式分量作为高频重建中的源信号。在得到高频重建的源信号后,再利用模式分量能量进行调整;最后通过源滤波器模型合成高频信号。实验结果表明,该方法与3GPP WB+和AVS P10编码标准中的带宽扩展算法相比,编码质量得到了明显提升;与最新的3GPP EVS中的时域带宽扩展算法相比,主观音质相当,码率下降33.3%。

An Approach of Audio Bandwidth Extension via Empirical Mode Decomposition

Audio bandwidth expansion is a standard technique within contemporary audio codecs to efficiently coding high frequency signal with low bitrates. In existing methods, in most case, the high frequency signal is generated by a duplication of low frequency and a few parameters of high frequency. This method is derived from an assumption which the correlation is exists in the high and low frequency. However, the perception quality of coding will degraded significantly if the correlation becomes weak. For this issue, we propose a new approach to rebuild the high frequency signal by empirical mode decomposition. In particular, the high and frequency is decomposed into intrinsic mode function (IMF), and select strong IMF components of low frequency as source of high frequency when rebuilding. And then extracted the energy gain of IMF between high and low frequency, the high frequency signals are synthesized by a source-filter model. Our experiments show that the perception quality of coding is improved significantly compare with 3GPP WB+ and AVS P10, and is competitive compare with the latest 3GPP EVS with bitrates drop by 33.3%.