应用支持向量机和人工神经网络对大气次声信号 识别的初步实验

语音情感识别在许多领域具有重要研究价值,不同声学情感特征在使用不同分类器进行分类时,识别效果具有明显差异。与语音情感有关的声学特征包括谱特征、韵律学特征、音质特征。该文提出一种特征融合的方法,将3种声学特征中具有最好识别能力的特征进行融合:保留在实验中表现稳定且有较高识别率的谱特征的全部特征,提取韵律学、音质特征的相关统计量作为辅助特征融合于谱特征中。实验表明,该文所提出的融合特征在使用同一分类器进行分类时,识别率优于单一特征;当使用不同分类器时,融合特征依然具有较好的识别能力,且识别性能稳定,3个数据集上均有较好的识别率,基本实现跨数据集识别。

Experimental study on atmospheric infrasound signal recognition using SVM and ANN

Aiming at the problem that the signal extracted from infrasound station monitoring data by STA/LTA algorithm still contained noise, we made preliminary experimental studies on the machine learning method of support vector machine and neural network. We used a method of wavelet packet decomposition to reconstruct the signals, and extracted the energy characteristics from them. We also analyzed the methods to improve the recognition ability. The experimental results showed that the recognition ability of the two methods can be improved to an acceptable level by optimizing the model structure such as the training data set was small.