摘要： 讨论了时频分布以及距离测度在说话人确认中的应用。实验采用语音信号的时频分布作为说话人模型，并根据分布之间的距离测度进行确认判决。选择了多个核函数和距离测度进行比较，采用改进的Nelder-Mead 算法，对核函数的参数进行优化，显著提高了确认的正确率，并在实验中得到了验证。

关键词: 时频分布；说话人确认；距离测度；核优化

Abstract: This paper explores in the application of time-frequency(TF) distributions and distance measures to speaker verification. The approach uses the TF distributions of speech signals as the speaker’s models, and then decides according to the distance between those TF distributions. It selects several kernels and distance measures to compare their performances and optimizes kernel parameters by means of improved Nelder-Mead algorithm, which advance the verification true-rate remarkably and is verified by the experiment

Key words: Time-frequency distribution; Speaker verification; Distance measure; Kernel optimization