摘要
声音指纹技术现在已经广泛的应用到了歌曲搜索、乐曲识别、声音修复等各个领域,但其关键技术———音频降维技术仍存在分类效果不好、可靠性不高等问题。针对音频数据高维化存在较大随意性,提出了基于模式识别的音频数据高维化的最优方法。并在此基础上,提出了采用加权PCA方法作为声音指纹的降维技术,不仅分类效果大为明显,且由于方法还保持了线性方法的简单性,保证了大批量处理数据成为可能。
Audio fingerprint technology has been widely used in the music searching, melody identification, and sound restoration. However, dimensionality reduction, the key to audio fingerprint technology, still cannot achieve satisfactory classification and reliability. Firstly, this paper introduced an optimal audio-dimensionality-segment method based on pattern recognition. Secondly, weighted PCA（Principal Component Analysis） was suggested as the kernel dimensionality reduction technology in audio fingerprint processing. This method not only enhances the classification of music data, but also keeps the merits of linear dimensionality reduction, simplicity and fast computation, which makes the heavy-data-precessing become feasible.
出处
《计算机应用》
CSCD
北大核心
2006年第9期2250-2254,共5页
journal of Computer Applications
关键词
加权主成分分析
声音指纹
线性降维
weighted PCA（Principal Component Analysis）
audio fingerprint
linear dimensionality reduction
