| With the rapid development of the Internet and information technology, the voiceprint recognition technologyhas widespread application value and prospect needsin the financial, securities, social security, e-commerce, banking and other remote customer service confirmed,and the specific identity of public security, military security sector of automatic testing and certification, it is important to explore the direction of the sound signal processing and biometric information detection and identification of areas in whole world,in recent decades, researchs has made significant progress in this field, but because of the speaker personality susceptible to external factors and the complex nature of the specific physical environment, which highlights the bottleneck effect gradually, so it can be more effective in voice detected complex background information and to finding out more robust feature extraction algorithm for improving the recognition rate of the system has a very important significance.Voiceprint identification system is inhighly complex noisebackground, the voice is detected and further feature extraction, and then through the modeling features extracted after analysis and processing, the final speaker recognition and identification, it is one ofbiometric ways to verify the identity information. The Paper research speech endpoint detection method and feature extraction algorithms to identify efficiency of Voiceprint identification system,The main work is as follows.First of all, in the sound preprocessing stage, we propose two different voice signal endpoint detection method based on a noisy environment. depending on the complexity of the different level of background noise ratio,wecorrespond tochoice using spectral entropy algorithm and dual-threshold endpoint detection algorithms based on zero-rate and short-term energy. Experimental results show that dual-threshold endpoint detection algorithm results are satisfactory the in high SNR-based short-term energy and zero-crossing rate, and endpoint detection algorithm based on spectral entropy is better inbackground under low SNR.Secondly, in the feature extraction stage, calculating pitch cepstrum parameters usecepstrum, and converting the voice signal power spectrum to Mel Cepstral Coefficients (MFCC)by Mel filter bank, then using the improved feature extraction algorithm compose two parameters as a voiceprint acoustic pattern, and both of them were conducted simulation experiments.Finally, in voiceprint identification stage, recognition algorithm with noise characteristics (SEMG) algorithm proposed firstly, that using the spectrum entropy endpoint detection algorithmdetect the speech signal endpoint, and then using the improved feature extraction algorithmconduct feature extraction, finally establishinga Gaussian mixture model (GMM) for everyspeaker, and validated SEMG algorithm to achieve the desired results by experiments. |
