中国科学院机构知识库网格
Chinese Academy of Sciences Institutional Repositories Grid
音乐音频的节奏识别

文献类型:学位论文

作者杜云峰
学位类别博士
答辩日期2007-06-01
授予单位中国科学院声学研究所
授予地点声学研究所
关键词节奏识别 激励检测 基频分析
其他题名Tempo Tracking for Musical Audio
学位专业信号与信息处理
中文摘要随着数字音乐媒体在网络和个人PC 间的大量交互,如何高效方便地组织与处理这些数字音乐信息已成为现今多媒体应用和计算机技术领域的热点问题。 节奏识别技术既是由此应运而生—用户可以通过计算机自动识别数字音乐文件的节奏值(BPM),进而对音乐进行节奏、流派、情感的分类处理与应用等。 本文对节奏识别技术进行了研究且实现了一个完整的节奏识别系统,具体内容与成果如下: • 研究了音频激励检测技术(节奏识别系统的前端模块),提出了一种新的基于非零标准差计算门限准则的峰值提取算法。实现的激励检测算法在国际音乐信息检索评测比赛(MIREX2006)中获得了4 项指标第一名(共9 项指标),其中团体名次第二名,复杂音乐类型第一名。 • 研究了基频分析技术(节奏识别系统中端模块),提出了真实高分辨率分析(不再按FFT 谱线分辨率进行规整)的处理方法。实现的基频分析算法在适用范围与性能方面均优于实验室以前版本的算法。 • 研究了时频矩阵后处理技术(节奏识别系统后端模块),提出了一种新的基于上下文叠加的后处理方法,算法的“性能-复杂度比”明显优于传统的动态规划后处理方法。 • 实现的整个节奏识别系统在ISMIR2004 节奏检测比赛数据库上的实验性能排名第二且具有较高的运行效率(与其他12 个参赛算法模拟比较)。
英文摘要Since the inter-change of digital musical media among internet and PCs has become more and more, how to organize and process this kind of digital musical information efficiently and conveniently has become a hot problem in nowadays. Therefore, the technique of tempo tracking has been proposed — users could utilize computer to automatically detect and track the tempo value (BPM) of any digital musical file, furthermore, to implement the applications of rhythm, genre, emotion classification and etc. This diploma thesis is oriented to research on tempo tracking, and introduces a complete framework of tempo tracking system. Detail contents and results as follows: • Research on audio onset detection (front-end of tempo tracking system), it proposed a novel peak-picking algorithm based on a non-zero standard deviation threshold. The proposed onset detection algorithm ranked 1st in 4 sub-tasks (total 9 sub-tasks), especially 1st in the sub-task on the complex music class, in the Audio Onset Detection Contest of MIREX2006. • Research on pitch tracking (mid-component of tempo tracking system), it proposed a kind of real high-resolution processing method (not based on the resolution of FFT bins). The proposed pitch tracking algorithm achieved better performance and larger processing range than the lab’s previous algorithms. • Research on post-processing for time-frequency matrix (back-end of tempo tracking system), it proposed a novel post-processing algorithm “contextbased accumulation”. The cost-performance of the proposed post-processing algorithm is much better than the traditional DP-based post-processing method. • The proposed tempo tracking system achieved high computational efficiency and ranked 2nd performance on the experimental dataset of ISMIR2004 Tempo Induction Contest (compared with other 12 participated algorithms).
语种中文
公开日期2011-05-07
页码81
源URL[http://159.226.59.140/handle/311008/242]  
专题声学研究所_声学所博硕士学位论文_1981-2009博硕士学位论文
推荐引用方式
GB/T 7714
杜云峰. 音乐音频的节奏识别[D]. 声学研究所. 中国科学院声学研究所. 2007.

入库方式: OAI收割

来源:声学研究所

浏览0
下载0
收藏0
其他版本

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。