Combining standard and throat microphones for robust speech recognition

The effect of bone conduction microphone locations on speech intelligibility and sound quality

“…The throat microphone is attached to the neck directly on top of the skin covering the vocal cords and is not as susceptible to ambient noise as an air microphone (Graciarena et al, 2003). However, the transmission capabilities of throat microphones are inferior to air microphones, they frequently produce user discomfort, and their output is very susceptible to changes in microphone location.…”

Recent developments in signal processing for digital hearing aids

“…Recently, boneconductive microphones have received increasing attention and have found uses in many aspects of voice signal processing [8], [9]. Unlike regular airconductive microphones, bone-conductive microphones receive vibrations only from the bones of the human talker and from nearby air.…”

Automatic speech segmentation using throat-acoustic correlation coefficients

“…It is possible to more precisely consider the special features of voice excitation process using the additional information which re ects the given process dynamics more accurately. A throat microphone signal can be used as a similar additional informational component [15][16][17][18][19][20][21][22][23][24][25][26][27][28]. In its turn, though a throat microphone signal reects the features of vocal fold oscillations more precisely in comparison with a speech signal, additionally the question arises concerning the identi cation of informative features on the basis of the synchronous analysis of a speech and a throat microphone signal which can be used in the speech signal segmentation process.…”

“…In [15][16][17][18] the question of combining an acoustic microphone with a throat microphone or an osteophone for better speech intelligibility, as well as for robust speech recognition in noisy environment was considered. For example, in [15] the author proposed a special algorithm for mapping a noisy time series of features from an acoustic microphone to the corresponding sequence from a throat microphone, which is used as an extended feature vector with following noise component ltration in the microphone mel-frequency cepstral coe cients. In the process of continuous noisy speech recognition, this algorithm has shown a signi cant improvement of e ectiveness in comparison with the approach of using only one acoustic microphone.…”