基于时空域转换的音频信号分析与识别Analysis and - 数码设计.PDF

第5 卷 第2 期 数 码 设 计 Vol.5 No.2 2016 年10 月 PEAK DATA SCIENCE October 2016 文章编号：1672-9129(2016)02-0041-06 基于时空域转换的音频信号分析与识别 * 刘雨青 ，刘艳芳 (龙岩学院 信息工程学院，福建省龙岩市 364000) (*通信作者电子邮箱：lyqfjnu@163.com) 摘要：音频信号的识别是实现计算机自动谱曲的基础，在音乐的创作中有很重要的实用价值。 本文通过对采集的音 频信号进行时域分析和频域分析，实现音频信号端点检测和音符识别。首先，运用短时能量和短时过零率两个时域 特征对音频信号进行端点检测和单音符分割；其次，通过频域分析，运用小波分解和 Gabor 变换对分割出的单音信 号进行时-频转换，去除泛音干扰分量，识别单音信号对应的基音频率；最后，将识别出的基音频率匹配到对应的乐 音音符。实验结果表明，该方法识别准确率较高，误差较小。 关键词：音频识别；单音分割；时频转换；短时能量；短时过零率；特征提取 中图分类号：TP391.42 文献标志码: A Analysis and Recognition of Audio Signal Based on Time-Frequency Transformation * LIU Yuqing , LIU Yanfang (Institute of Information Engineering, Longyan University, Longyan Fujian 364000, China) Abstract ：The recognition of audio signal is the basis of automatic-composing music in computer, thus it has great application value in music creation. In this paper, the time-domain analysis and frequency-domain analysis are carried out for the collected audio signal to realized endpoint detection and notes recognition. Firstly, short-time energy and short-time zero-crossing rate are used in endpoint detection and single note segmentation; secondly, through frequency-domain analysis, the time-frequency transformation of the segmented single note is performed by wavelet decomposition and Gabor transformation. After removing the harmonic interference, the pitch frequency which is corresponding to the single tone is r