基于文字符号的检索体系,对于音乐材料效果不佳,基于内容的音乐信息检索会成为今后发展的关键.针对音乐检索技术的核心--音频特征描述和特征匹配,本文以音高和相对节奏为音乐作品的关键特征,发展了非毗邻层连接的前馈神经网络结构,给出了误差反传训练算法的分类器,并进行了试验研究.结果表明,非毗邻层连接的前馈神经网络结构有优越的识别性能和极快的收敛速度.%The search system based on letter sign has bad effects on music material. Search of music information based on contents becomes the key of development from now on. We aim at the core of music search technique: audio frequency characteristic description and characteristic match. This paper takes the pitch and the opposite rhythm as the key characteristics of music work and develops the front feedback nerve network structure of non-adjacent layer conjunction, and gives the machine that the error margin is inverted spread the classification of training the calculate way and carries on to experiment a research. The result indicates that the front feedback nerve network structure of non-adjacent layer conjunction has predominant function on identifying and splitting velocity on convergence.
展开▼
