基于MFCC和双重GMM的鸟类识别方法  

Bird recognition based on MFCC and dual-GMM

作　　者：王恩泽[1] 何东健[1] 

机构地区：[1]西北农林科技大学机械与电子工程学院,陕西杨凌712100

出　　处：《计算机工程与设计》2014年第5期1868-1871,F0003,共5页Computer Engineering and Design

基　　金：西北农林科技大学校长基金项目(A213020901)

摘　　要：针对鸟类鸣声信号变化丰富和复杂的特点,提出一种基于MFCC和鸣叫、鸣唱声GMM模型的鸟类识别方法。该方法拟采用将鸟鸣声分为鸟叫声和鸟唱声的策略,分别提取其特征参数MFCC,提出双重GMM模型进行训练和识别。用8种鸟的鸣叫声和鸣唱声1077个样本进行实验,实验结果表明,双重GMM模型的识别率达到90%以上,与单一鸣声模型相比具有更高的识别率。In accordance with the flexibility and the complexity of bird chirps, a bird recognition method is raised on the basis of MFCC and GMM model. By classifying bird chirps into call and sing, and extracting their feature parameter MFCC respectively, a dual-GMM model is put forward for training and recognizing. Experiments are carried out with 1077 bird chirp samples of 8 kinds of birds. Results show that the recognition rate of dual-GMM model is over 90%, much higher than that of the single vocal model.

关 键 词：鸟类识别 梅尔倒谱系数 鸣叫 鸣唱 双重高斯混合模型 

分 类 号：TN912.34[电子电信—通信与信息系统]