基于语音信号时变特性的说话人辨认

在平均Mel倒谱基础上提取随时间变化的特征频率，由此得到了由各个语音信号特征频率倒谱值序列构成的时间序列。运用时间序列预处理和数理统计的方法，分离时间序列的趋势量和波动量。波动量是零均值自协方差非平稳的时间序列，利用满阶时变参数自回归TVPAR（Time—Varying Parameter Autoregressive）模型对波动量序列进行分析，进一步提取说话人语音信号的特征参数。在波动量序列和用满阶TVPAR模型分析的基础上分别进行说话人识别。实验表明，用满阶TVPAR模型进行识别，识别率比波动量序列上的识别率有较大提高，一个特征频率上平均识别率达到99．68％，取两个特征频率时达到100％。

Speaker Identification on the base of

Time-varying characteristic frequency was extracted from the average Mel cepstrum, and the cepstrum value series of characteristic frequency were gained. The deterministic and stochastic parts of the time series were separated by use of time series pretreatment and statistical methods. As zero mean autocovariance nonstationary time series, the stochastic parts were analyzed by the full order TVPAR（Time-Varying Parameter Autoregressive）model, and the characteristic parameters were extracted from speech signals of the speaker. Then the speech signals were recognized on the stochastic parts of the time series and analysis with the full order TVPAR model. The experimental results manifest that the recognition rate obtained by full order TVPAR model are higher than only on stochastic parts of the time series, with one or two characteristic frequencies, the average recognition rate reaches 99.68% and 100% respectively.