歌唱人声分离算法改进研究  

Research in the Improvement of Singing Voice Separation Algorithm

作　　者：汤政[1] 

机构地区：[1]北京邮电大学国际学院,北京102209

出　　处：《电声技术》2013年第12期61-65,共5页Audio Engineering

摘　　要：歌唱人声分离具有广阔的市场应用前景,但是它的真正实现需要解决目前速率低和硬件要求高的难题。本算法综合考虑乐音信号和人声信号的特点,对过去的算法进行了一些修改,使主旋律提取和人声分离的各个步骤相互补偿。在此基础上,本算法还采取了结构复用、减少迭代和去除功能的思路,提升运行效率。并在MIR-1K数据库样本上进行计算机仿真实验,证实本算法在性能上达到了预期目的。Singing voice separation has a promising marketing prospect for application, but the problems, including low speed and high hardware requirement, need to be solved before it can be realized. Considering the characteristics of music signal and vocal signal, the traditional algorithm is modified. Each procedure in melody extraction and vocal separation com- pensates each other. Moreover, this algorithm is optimized aiming at improving efficiency by reusing components, reducing iterations and eliminating functions. After computer simulation on MIR-1 K dataset, it is verified that the algorithm＇ s im- provement in performance achieves the desired effect.

关 键 词：人声源分离 效率提升 计算听觉场景分析 计算机仿真 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]