中文摘要：

为分析拉萨方言语音差异性，通过采集语音样本，获取其语音特征向量，并将特征向量代入AP聚类算法，结果认为拉萨方言语音存在较大的个体差异，可将覆盖率较高聚类簇作为拉萨藏语方言规范语音．运用现代科学定量研究藏语方言语音特征，以较高能量的语音帧作为语音特征量，帧能量比作为语音特征片段的起止点，数据选取上具有一定的代表性，数据代入传统的AP聚类算法，进而分析程序的运算结果，实验过程一定程度上体现了定性与定量相结合的原则，验证了算法的收敛性和鲁棒性．

英文摘要：

To analyze differences in Lhasa dialect speech, we collected voice samples to get their voice feature vectors, and then generation the feature vector into the AP clustering algorithm, conclusion shows that there are individual differences in Lhasa dialect speech, which provided the higher coverage cluster serving as Lhasa Tibetan dialects specification voice. This paper studies the use of modern science in quantitative chaxacteristics of Tibetan dialects voice, then getting higher energy speech frames as a voice feature quantity, marking frame energy as a voice feature segments starting and ending point, select representative data into the traditional clustering algorithms AP, and then computing the results analysis program. The experiment embodied the principle of combining qualitative and quantitative, validated algorithm convergence and robustness.