Abstract:
Gaussian mixture model-universal background model (GMM-UBM) is the most commonly used model in speaker recognition technology; the model has achieved very good results in many ex-periments. In this design, the GMM-UBM model is used in the abnormal sound detection. First, we process the aircraft engine sound signal, second extract the MFCC characteristic parameters, then train UBM model and last obtain the GMM-UBM model by MAP adaptive algorithm. The ultimate goal of the test indicates that the method could optimize the recognition rate decline due to interference change.