A study on Speaker Identification Using Spectral Ratio between Air-Transmitted Speech Sounds and Bone-Transmitted Speech Sounds

Abstract

A bone-conductive microphone which transforms bone-transmitted speech sounds through skull into electronic signals has two advantages. One is to operate in hands free mode, and another is to be anti-noise. Therefore, it is expected to match the man-machine interface for wearable computers.Here, we examined and analyzed the possibility of a speaker identification system with the MFCC of air and bone-transmitted speech sounds to 5 Japanese vowels. This paper shows that the system using spectral ratio between air-transmitted speech sounds and bone-transmitted speech sounds gives better performance than the system using air-transmitted speech sounds.

Bone- and air-conduction speech combination method for speaker recognition

“…Some methods can combine air-and bone-conduction speech, such as a model-based method (Hershey et al, 2004;Tsuge et al, 2008Tsuge et al, , 2009) and a feature-based method (Mori et al, 2007;Tsuge et al, 2008Tsuge et al, , 2009.We propose three methods to fuse both types of speech. First, in 2.1, we propose a feature combination method, which combines both types of speech on the feature parameter space.…”

“…Finally, in 2.3, we propose a similarity score combination method, which calculates the decision score using the speaker model trained by the air-conduction speech and the speaker model trained by the bone-conduction speech. Mori et al's (2007) method is similar to our feature combination method. Whereas their method fuses two types of speech in the spectrum domain, our method fuses two types of speech in the feature vector domain.…”