摘要
为了实现猪只不同状态下声音的自动监测,试验采用声音识别技术,首先将猪只不同状态的声音信号进行双门限端点检测和预加重处理,然后通过大量试验对比,采用小波阈值法对声音信号进行去噪处理,并提取梅尔倒谱系数(MFCC)和一阶差分梅尔倒谱系数(ΔMFCC)作为描述特征,建立隐马尔可夫模型(HMM),最后对猪只不同状态的声音进行自动识别。结果表明:猪只状态识别精度较高,有助于提高自动监测系统的智能化判断能力。
In the intensive breeding, automatic recognition technology for pig states under the background of complex noise has an important sig- nificance for automatic breeding. To achieve automatic monitoring of voice from pigs under eight states, sound - recognition technology was used in the test. Firstly, the pig＇s sound signals in different states were used for double - threshold endpoint detection and pre - emphasis, and then were used for de - noising treatment using the wavelet threshold method through a lot of experimen comparisons. Mel - frequency cepstrum coef- ficient （MFCC） and first - order difference MFCC were extracted as descriptive characteristics to establish a hidden markov model（HMM）. Fi- nally, the pig＇s sounds in different states were automatically recognized. The results showed that the recognition accuracy for pig states was rela- tively higher, and it was helpful to raise the capacity for intelligent judgment in the automatic system.
出处
《黑龙江畜牧兽医》
CAS
北大核心
2016年第11期97-99,103,294,共5页
Heilongjiang Animal Science And veterinary Medicine
基金
"863"国家高技术研究发展计划项目(2013AA102306)
关键词
猪只状态检测
声音识别
隐马尔可夫模型(HMM)
梅尔倒谱系数(MFCC)
一阶差分梅尔倒谱系数
端点检测
预加重
小波阈值去噪
pig state detection
sound recognition
hidden Markov model （HMM）
Mel -frequency cepstrum coefficient（ MFCC ）
Mel -ceps- trum coefficient of first - order difference
endpoint detection
pre - emphasis
wavelet threshold denoising
