基于分形维数和模糊RBF神经网络的语音端点检测

简单介绍了分形维数的概念及模糊RBF神经网络的结构。利用分形维数在噪声情况下作为语音端点检测参数的优越性，组合幅度熵、帧能量及过零率作为模糊神经网络的输入参数进行语音信号端点检测。用连续语音进行非正式测试，实验证明该方法避免了选取阈值这一难点，在噪声情况下仍具有较高检测准确率。

The Endpoint Detection Algorithm of Speech based on Fractal Dimension and Fuzzy RBF Neuron Network

This paper briefly introduces the concept of the fractal dimension and the structure of the fuzzy RBF neural networks.By making use of the advantage of the fractal dimension in the noise situation as the parameters of voice signal endpoint detection,and combining amplitude entropy,frame energy and the zero crossing rate as fuzzy neural network input parameters for voice signal endpoint detection.With the informal test of continuous speech,the experiments show that the method can avoid the difficult to select threshold value,which also have higher detection accuracy rate in the noise situation