Estimating Transfer Function from Air to bone conduction using singing voice

Singers often notice a disparity between how they perceive their own voice while they are singing and how they perceive their voice when hearing a recorded playback. This research aims to produce a digital filter which will take a sound sung by a singer and render it to a processed sound which is a close match to that heard by the singer during performance. One reason for the disparity arises from the fact that the skull conveys sound to ears when people speak or sing, and then the sound that is delivered blends with the air-conducted ones at the cochlea. In this paper, we investigate the spectral difference between the air-conducted sound and the boneconducted sound, and estimate the transfer function from air conduction to bone conduction by signal processing methods. Then we transform the original recording by air transmission using the filter coefficients of the transfer function and compare the filtered sound to the original recording by bone transmission.