Abstract: AVS-P10 is Chinese first mobile audio codec standard. To solve the problem of low quality because of low accuracy in open-loop mode selection of the standard, this paper proposes an open-loop mode selection method based on random forest which regards closed-loop mode choice as the benchmark to improve mode selection accuracy of open-loop mode. To keep low computational complexity and high accuracy, the scale of random forest is reduced by two-step feature selection algorithm. Both of the results of objective and subjective testing show that the accuracy rate of mode selection increases dramatically. The quality of audio encoded by the open-loop selection using this method is much better than that by open-loop selection of AVS-P10 and is almost same as that by close-loop selection.

Key words: random forest, Audio Video coding Standard-Part 10（AVS-P10）, mode selection, feature selection, signal classification

摘要： AVS-P10是我国第一部移动环境下的音频编解码标准。针对AVS-P10开环方式下编码模式选择算法准确率不高导致编码质量不佳的问题，提出基于随机森林的开环编码模式选择算法，通过拟合闭环方式下的编码模式选择结果，以提升开环方式下的编码模式选择准确度。为了降低随机森林计算复杂度并保持高准确度，采用两次特征选择并结合随机森林的特性选择特征。主客观测试结果表明，所提方法将开环方式下的编码模式选择准确率大幅提升；采用所提方法的开环编码质量明显优于AVS-P10开环编码质量，与AVS-P10闭环编码质量相当。

关键词: 随机森林, 先进音视频编码标准第10部分（AVS-P10）, 模式选择, 特征选择, 信号分类