Accuracy improvement of speaker authentication in noisy environments using bone-conducted speech

Speaker authentication has received much attention recently in biometrics authentication technology since it needs only microphones and telephone networks which are standard equipments for voice communication and widely used. However, there are problems where speaker authentication systems are sensitive to noise. Thus, Iijima et al. attempted to utilize bone-conducted speech that is robust against noise, instead of airconducted speech. Additionally, a speech transformation method was suggested for improving the authentication accuracy in highly noisy environments. In this paper, we analyze the problem of Iijima's speech transformation method and propose two new speech transformation methods. Experiments verify that the proposed speech transformation methods are effective in highly noisy environments.