Text-independent Speaker Verification System And Its Application

In this paper we mainly discuss a speaker verification system based on the UBM-GMM model. Based on the implementation of the system, we build a new speech database, and add the double-bounds endpoint progress which is based on the short-time energy and zero-crossing, and we get a fairly good verification result after that work. And then, for the runtime of the system is a little too long because of the large amount of data, we give a modification that we just extract the parameter to the most effective part of the speech, and we can see from the test result that the runtime after modification has declined to one third of former one, and the EER of the system does not change much.In this paper we first discuss the background, usage and current developing condition of the speaker verification system, and give the shortages of it up to now. Then give the introduction of the speaker verification system by its flow process. In this paper we use the DET curve to show the EER of the system. For the SNR of the former speech database is too low, we build a new database, and add the progress of the double-bounds end-point, so we can get a fairly good result.But we can also see in the test that because of the large amount data, the runtime of the system is a little too long, especially if we use it in the real-time condition. So about to this shortage we give a modification. We first get the score of every frame, and get the part with high score. Then we analyze the short-time energy and zero-crossing of this part, to find the differences of it to other parts, and we can see that we select this part by the short-time energy and zero-crossing. Then we just extract the MFCC to this part and give s test, we can see that the run-time of speaker verification system has declined to a third of the former one, and the EER of the system does not decline, and even get a little rising.At last we give a discussion of the pervasive computing system, and introduce a mandarin e-learning system based on it, and introduce the usage of the speaker verification system in the E-Learning.