Speaker Identification Method Using Bone Conduction and Throat Microphones

Abstract

In recent years, the number of workers in Japan’s construction industry has been decreasing, and the working environment at construction sites and other facilities is being improved. For example, the use of professional headsets equipped with bone conduction microphones, which enable clear voice communication even under noise, is expanding. Therefore, we aim to improve the performance of speaker identification using bone conduction microphones. In this study, we propose a speaker identification method using the difference of spectrum between bone conduction and throat speech. In the proposed method, Mel log power spectrum of bone conduction and throat speech is calculated. Then, the difference of spectrums is calculated and SVM is used to identify the speaker. The proposed method using the difference of spectrum between bone conduction and throat speech achieves the highest identification rate, which confirms the effectiveness of the difference of spectrum between bone conduction and throat speech.