自适应并行模型组合的鲁棒语音身份识别算法

Robust Speaker Identification Algorithm Based on Adaptive Parallel Model Combination

由于环境噪声的影响,实际应用中说话人识别系统性能会出现急剧下降。提出了一种基于高斯混合模型-通用背景模型和自适应并行模型组合的鲁棒性语音身份识别方法。自适应并行模型组合是一种噪声鲁棒性的特征补偿算法,能够有效减少训练环境与测试环境之间的不匹配现象,从而提高系统识别准确率和抗噪性能。首先,算法从测试语音中估计出噪声特征,然后用一个单高斯模型对噪声特征进行拟合得到噪声均值和协方差。最后,根据得出的噪声均值和协方差,调整训练好的高斯混合模型均值向量和协方差矩阵,使其尽可能地匹配测试环境。实验结果表明,该方法可以准确地重构干净语音的高斯混合模型参数,并且能够显著提高说话人识别的准确率,特别是在低信噪比情况下。

The performance of speaker recognition systems degrade rapidly in real applications due to environmental noise.This paper proposes a robust speaker recognition method based on Gaussian Mixture Model-Universal Background Model(GMM-UBM)and adaptive parallel model combination(APMC).APMC feature compensation algorithm,which is robust to noise,can effectively reduce the mismatch between training environment and testing environment so as to improve the recognition accuracy and anti-noise performance.Firstly,automatically estimating noise feature from test speech.Secondly,using a single Gaussian model to fit the feature,then getting the mean and covariance of noise feature.Finally,according to the mean and covariance of noise from the second step,the mean vectors and covariance matrices of the training GMM are transformed to the testing condition by this method as far as possible.The experimental results indicate that the proposed method can reconstruct the clean speech GMM parameters more accurately.Also,this method can significantly improve the speaker identification accuracy,especially in low SNR.