基于非线性音频特征分类的频带扩展方法

提出了一种基于非线性音频分类的频带扩展方法，即利用递归图和定量递归分析将音频信号的时间序列分成4类，并分别采用4种方法恢复高频频谱细节，最终利用高斯混合模型和基于软判决的码书映射调整频谱包络和能量增益。主客观测试表明，该方法优于传统的盲目式频带扩展方法，且应用到ITU-T G.722.1编解码器时，音频质量优于同码率下的G.722.1C编解码器。

Bandwidth extension method based on nonlinear audio characteristics classification

A bandwidth extension method based on audio classification was proposed. Time series of audio signals were classified into four types based on recurrence plot and recurrence quantification analysis, and the fine spectrums were recovered by taking advantage of four methods respectively. In addition, the spectrum envelope and energy gain were adjusted by Gaussian mixture model and codebook mapping on the basis of soft decision respectively. Subjective and objective testing results indicate that the proposed method has good quality compared with conventional blind bandwidth extension methods, and the performance of ITU-T G.722.1 codec with the proposed algorithm is better than that of G.722.1C codec at the same bit rate.