噪声背景下基于多模板矢量量化的与文本无关的话者辩识

A Method Based On Multiple Vector Quantization For Robust Text-independent Speaker Identification

在话者辨识系统的实际应用中，导致系统识别率下降的根本原因是噪声的影响，它使得测试与训练条件不一致。本文针对实际环境中常见的加性背景噪声，提出了利用加入不同类型、不同信噪比噪声的含噪语音进行训练说话人的模型，每个说话人具有多个模板。实验结果表明，这种方法能够有效的提高系统的鲁棒性。文中还讨论了距离加权方法在话者辨识中的应用。

The system performance will be degraded rapidly if the testing conditions do not accord with the training ones in speaker identification system. this paper introduces an approach for training speaker codebooks using multiple session training speech samples with every speaker having mutiple codebooks. These different session. training speech samples are obtained through adding several kinds of noise with different SNR to the original clear speech. Experiments show that this technique the robustness of system effectively