An Efficient Feature Selection Method for Speaker Recognition

In this paper, a new feature selection method for speaker recognition is proposed to keep the high quality speech frames for speaker modelling and to remove noisy and corrupted speech frames. In order to obtain robust voice activity detection in variety of acoustic conditions, the spectral subtraction algorithm is adopted to estimate the frame power. An energy based frame selection algorithm is then applied to indicate the speech activity at the frame level. The eigenchannel based GMM-UBM speaker recognition system is used to evaluate this proposed method. The experiments are conducted on the 2006 NIST Speaker Recognition Evaluation core test condition (telephone channel) as well as microphone channel test condition. It demonstrates that this approach can provide an efficient way to select high quality speech frames in the noisy environment for speaker recognition.