Learning state-dependent stream weights for multi-codebook HMM speech recognition systems | IEEE Conference Publication | IEEE Xplore