Abstract: In text-independent speaker identification，robustness is very important especially for phone speech.Score normalization which is used for speaker verification is introduced into speaker identification for the first time.Through score normalization，the output scores are more reasonable for speaker identification.Experiments on text-independent GMM speaker identification in NIST’03 1spk data show noticeable improvement.

摘要： 在文本无关的说话人辨识中，为了提高系统在电话语音条件下的鲁棒性，提出了将说话人确认中常用的评分规整手段用于说话人辨识中，即对测试语音通过不同话者模型的评分分别进行评分规整，为测试语音选取最接近的话者模型作为系统识别输出，有效地提高了系统性能。在NIST’03 1spk数据库上的说话人辨识实验表明了评分规整技术对说话人辨识的有效性。