基于PCNN的语谱图特征提取在说话人识别中的应用

Research of Feature Extraction from Spectrogram Based on Pulse Coupled Neural Network in Speaker Recognition

该文首次提出了一种将有生物视觉依据的人工神经网络——脉冲耦合神经网络(PulseCoupledNeuralNetwork,以下简称为PCNN)用于说话人识别领域的语谱图特征提取的新方法。该方法将语谱图输入到PCNN后得到输出图像的时间序列及其熵序列作为说话人语音的特征,利用它的不变性实现说话人识别。实验结果表明,该方法可以快速有效地进行说话人识别。该文将PCNN引入到语音识别的应用研究中,开拓了信号处理中两个极为重要的部分———语音信号处理和图像信号处理结合的新领域,同时对于PCNN的理论研究和实际应用具有非常重要的现实意义。

This paper proposes an innovative method of feature extraction of Spectrogram using Pulse Coupled Neural Network(PCNN) in speaker recognition for the first time.Using the invariable characteristics of PCNN time series and entropy series of Spectrogram,people can extract the feature of speaker's speech and recognize the speakers rapidly and effectively.The validity of this method is well verified by authors' experiments.It is significant for theoretical research and application of PCNN to bring PCNN into speech recognition.At the same time this paper envisages a combination of two important parts of signal processing-speech and image processing.