Speaker Recognition Research In Noisy Environment

Speech Signal Processing technique is indispensable in information society. Speaker Recognition (SR) is one of important research branches of speech signal processing field. SR is also called voiceprint recognition. Its task is to identify or verify the identity of the speaker by analyzing and recognizing specific information extracted from speech waves of the speakers. It integrates theoretic achievements of many disciplines, such as Acoustics, Phonetics, Linguistics, Physiology, Digital Processing, Pattern Recognition and Artificial Intelligence, etc. SR technique has a broad application prospects in judicial verification, confidential secure and e-business fields.SR system is mainly composed of feature selection and pattern recognition. The purpose of feature selection is to extract data which reflects the speakers'personal characteristic from many original speech signals. The goal of pattern recognition is to design the valid classifier. This project uses Mel-Frequency Cepstral Coefficient (MFCC) difference MFCC (ΔMFCC) to feature extract and adopts a traditional VQ with good recognition performance to finish the Recognition model and pattern matching. At last, a test platform of text-dependent and close set speaker identification system is established under the Matlab software. Experiments have tested on a speech database contained twenty speakers. When in clean voice environment, we can obtain a much higher recognition ratio.Though there is a long history of SR's research, there are some problems to be solved. And one of the problems is to recognize the speaker under noisy environment. A SR system with high recognition performance in relatively clean environments becomes deficient in the noisy environments. The experiments showed the same result, which is the lower the signal-to-noise ratio, the lower the system recognition ratio.SR should not only remain in the laboratory. To put SR into practical application, SR system needs to have very good robustness to noises under all different noisy environments and different Signal -to -Noise conditions. The project takes Gauss White Noise environment into consideration. We select 10dB 15dB 25dB 30dB 40dB noisy signals. In order to improve SR robustness under noisy environment, Wavelet de-noising method is used to suppress signal noises. The test result shows that the system recognition ratio is improved after the de-noising process. It makes the designed SR system project have certain practical meaning.There is no doubt that finding and using efficient de-noising method is important and useful to improve the performance of this system. It's also one of the most important research directions of my work in the future.