A multi-SNR subband model for speaker identification under noisy environments