Speech Language Classification Algorithm Based on Multiple-instance Learning with Negative Sample Cleaning

Speech language classification is an important direction in the field of machine learning. In order to further improve the effect of speech classification, this paper propose a new speech language classification algorithm based on multiple-instance learning with negative instance cleaning, namely NUMI-Biased SVM algorithm. Firstly, extract MFCC features from speech files, regarding each file as an instance bag, and every speech frame as an instance. Secondly, adopt PU learning algorithm to clean negative instances in positive bags to reduce their impact on classification. Finally, multiple-instance learning algorithm is used to classify the speech files. Experiments show that, compared with the SVM algorithm and multiple-instance SVM algorithm, NUMI-Biased-SVM has achieved significant improvement in classification performance.