Abstract: Aiming at problem of packets loss, this paper proposes a method of lost packets compensation based on Lagrangian interpolation and a new classifier GMM-DM. The algorithm improves performance of GMM classifier when training data is inadequate due to the packets loss during transportation. Experiments show that compensation based on Lagrangian interpolation and GMM-DM new classifier could obtain better results than traditional methods when the ratio of lost packets is relatively high.

Key words: speaker recognition, lost packets compensation, GMM-DM

摘要： 针对说话人语音数据在网络传输过程中的丢失问题，该文提出了一种基于Lagrangian插值的分组恢复方法，评估了丢失帧的实际位置，效果良好，改进了GMM识别算法，分析了一种基于GMM-DM的识别算法，克服了数据丢失对系统识别率的影响。实验结果表明，Lagrangian插值分组恢复方法和GMM-DM识别算法，在丢包率比较大时，可以减小丢帧而造成的负面影响，在训练数据不充分时，提高了系统的识别率。

关键词: 说话人识别, 丢包补偿, GMM-DM

CLC Number: 

• TP18