摘要
传统的乐器识别方法采用的是树型分类方法,这种方法分类过程比较繁琐,而且精度不高。该文把话者识别的方法应用到乐器识别之中,采用模式识别的方法实现对乐器的识别。采用MFCC系数和它的一阶导数作为音品的声学特征,分别对6种管弦乐器建立高斯混合模型。在识别过程中,首先假设各乐器的先验概率相同,根据高斯混合模型得出的后验概率确定待识别乐器所属的种类。实验表明这种识别方法十分有效,取得了较高的识别精度。
The traditional instrument recognition method adopts binary-tree classifying method. The process of this method is trivial and inaccurate. This paper applies speaker recognition methods into instrument recognition. The pattern recognition is utilized to implement instrument recognition. The MFCC coefficient and its derivative are taking as the acoustic features. A GMM model is constructed for each instrument set. In the process of recognition, the prior probability is supposed to be the same, the posterior probability is calculated according to GMM, and then the instrument class is determined. The experiment shows that this method is quite efficient and has better precision.
出处
《计算机工程》
CAS
CSCD
北大核心
2004年第18期133-134,173,共3页
Computer Engineering
关键词
高斯混合模型
乐器识别
话者识别
Gauss mixture model
Instrument recognition
Speaker recognition
