歌唱人声分离算法改进研究 research in the improvement of singing voice separation algorithm.pdfVIP

歌唱人声分离算法改进研究 ·算法研究· 汤政 (北京邮电大学国际学院，北京102209) 【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一1K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。 【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 【中图分类号】TP391．9 【文献标志码】A Researchinthe of Voice ImprovementSingingSeparationAlgorithm TANG Zheng ofPostsand (International University 102209，China) School，Beijing Telecommunications，Beijing voice hasa for the low 【Abstract】Singing problems，including separationpromisingmarketingprospectapplication，but tobe it the ofmusic and hardware solvedbeforecanbe characteristics speedhish requirement，need realized．Considering andvocal traditional ismodified．Eachin extractionandvocal com- signal signal，the algorithm proceduremelody separation each is at pensatesother．Moreover，thisalgorithmoptimizedaimingimprovingefficiencybymusingcomponents，reducing and functions．After K im— iterations simulationonMIR-1dataset．itisverifiedthatthe eliminating computer algorithm’s in achievesthedesiredeffect． performance provement source scene simulation 【Keywords】vocal separation；efficiencyimprovement；computationalauditoryanalysis；computer