一种基于K-SVD的说话人识别方法

为了充分提取语音中的个人特征信息,类比矢量量化,提出了一种基于K-均值奇异值分解(K-SVD)的说话人识别方法。利用K-SVD训练得到的字典可较好地保存语音信号中的个人特征信息。利用这一特性,通过K-SVD从训练数据中提取包含说话人个人特征信息的字典,利用该字典实现说话人识别。相对于传统方法,该方法能够更好地利用语音的稀疏性保存语音中的个人特征信息并减小重构误差。实验仿真结果表明,与基于矢量量化的说话人识别方法相比,该方法在多说话人的情况下具有更好的识别率,具有更高的实用价值。

Speaker recognition method based on K-SVD

In order to extract the personal characteristics, a speaker recognition method based on K-means Singular Value Decomposition （K-SVD） is proposed. The personal characteristics in voice can be well preserved in the dictionary trained from the K-SVD. With this feature, the dictionary which contains the personal characteristics is extracted from training data through the K-SVD algorithm. Then the trained dictionary is used for the speaker recognition. Compared to traditional methods, the personal characteristics in voice can be better preserved based on the proposed method through the sparse nature of voice and can reduce the reconstruction error. Experimental results show that the proposed method outperforms the VQ based methods for too many speakers in the view of recognition rate, so the proposed method has more practical value.