A de-noising algorithm for voice recognition with low SNR

摘要

Voice recognition technology is very important for prevention of identity crimes and cybercrimes because voice contains a lot of personal information. In this research, we propose a voice recognition system which consists of two stages of the feature extraction and identification. In feature extraction, noise directly affects the reason of feature extraction. Therefore, how to effectively remove noise is a special important study voice recognition system. Voice signal will be unavoidably affected by the interference of noise in the process of generation and transmission, thus resulting in a decline in the recognition rate of the system. When someone now enroll and use themselves voiceprint with noise to verify their identity by our presented system, the next time only use an app to access online services or phone, this security recognition system will help online system protect users account and reduce the chances of scammers accessing users account and information. The results of experiments, when SNR are -5, 0 and 5, the recognition rate of our proposed system can reach 80.1%, 86.2% and 90.4%, respectively. This research fits to the key approaches to prevent cybercrimes.