基于SVM和增强型PCP特征的和弦识别

Chord Recognition Based on Support Vector Machine and Enhanced PCP Feature

和弦识别是自动音乐标注的基础,在歌曲翻唱识别、音乐分割及音频匹配等领域具有重要作用。针对不同乐器之间相同和弦识别率较低的问题,提出一种基于瞬时频率提取音级轮廓(PCP)特征的改进算法。该算法结合音高频率倒谱系数,将增强型PCP特征作为新的和弦识别特征,把音频信号输入到节拍跟踪器,依据动态规划算法提取信号的节拍信息,计算音频信号每一个节拍内的增强型PCP特征,采用结构化支持向量机分类方法实现对音乐和弦的识别。实验结果表明,与传统PCP特征相比,采用增强型PCP特征的和弦识别率提高了2.5%～6.7%。

Chord recognition is the base of automatic music label, which plays an important role in the fields of song cover recognition, audio segmentation and audio matching etc. Among the different instruments, the recognition rate of the same chord is low. This paper proposes an improved chord recognition algorithm which combines the Pitch-frequency Cepstral Coefficients（PFCC） with Instantaneous-Frequency-based（IF） Pitch Class Profile（PCP） and uses the improved PCP as the new chord recognition feature. It inputs the audio signal into the beat tracker to extract the beat information of the signal which is based on dynamic programming algorithm, and calculates the improved PCP feature of the audio signal within each beat and realizes chord recognition by the structured Support Vector Machine（SVM）. Results show that the ratios of chord recognition increases by 2.5%~6.7%after using the improved PCP than using the traditional PCP.