基于小波倒谱系数和概率神经网络的取证说话人识别模型

取证说话人识别是一种利用犯罪现场留下的质疑语音来识别犯罪分子身份的技术。为了提高识别模型的抗噪能力，本论文提出了基于小波倒谱系数（WCC）和概率神经网络（PNN）的取证说话人识别模型。该模型包含WCC特征提取和PNN分类两个步骤。WCC对噪音不敏感，所以使得我们的模型有抗噪能力。PNN是一种高效的分类算法，从而提高了模型识别性能。实验表明，该模型以提高时间消耗为代价提高了识别率和抗噪能力。

Forensic Speaker Recognition Model using Wavelet Cepstral Coefficients and Probabilistic Neural Network

Forensic speaker recognition refers to recognize the identity of the criminals from the questioned speech obtained by the criminal scene. To improve the anti-noise ability of recognition model, we propose a forensic speaker recognition model based on wavelet cepstral coefficients(WCC) and probabilistic neural network(PNN). WCC is insensitive to noise, so feature vector obtained by it is robust in noise environment. Moreover, PNN is a effective classification algorithm, so it can improve the performance of our model. The experiments show that the proposed model obtains good performance in clear and noise environment at cost of increasing the time cost.