为了改善重音判决器的判决性能,引入多分类器准则,综合不同特征组合下独立判决器的优点,作出最终判定。针对单分类器的研究既包括单Gauss满阵模型,也包括多Gauss混合模型(GMM)。采用多分类器组合,利用各单分类器对训练样本和待测样本的决策信息,作出最终判定。实验表明,多分类器组合可以很好地结合不同特征组合组成的判决器的优点,提高识别率。采用多分类器组合可以在单Gauss满阵及混合Gauss元数不高的情况下将正确率提高到令人满意的数值(>80%),体现了很好的实用性。 In order to improve the judgment performance of accent judger, multi-classifier criteria are introduced to make the final judgment by synthesizing the advantages of independent verdict in different feature combinations. The research on single classifiers includes both single Gauss full matrix model and multi-Gauss mixture model (GMM). Multi-classifier combination, the use of single-classifier on the training sample and the decision-making sample information to make a final decision. Experiments show that the multi-classifier combination can well combine the advantages of the decision maker composed of different feature combinations to improve the recognition rate. The multi-classifier combination can improve the accuracy to a satisfactory value (> 80%) under the conditions of single Gaussian full matrices and mixed Gauss elements, which shows good practicability.