基于分形维数和模糊RBF神经网络的语音端点检测  

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

作　　者：张振红[1] 张雪英[1] 

机构地区：[1]太原理工大学信息工程学院,太原030024

出　　处：《电脑开发与应用》2008年第7期37-39,共3页Computer Development & Applications

基　　金：山西省自然科学基金资助项目(20051039)

摘　　要：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。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

关 键 词：分形维数 幅度熵 模糊神经网络 端点检测 

分 类 号：TN912.3[电子电信—通信与信息系统]