State-dependent mixture tying with variable codebook size for accented speech recognition | IEEE Conference Publication | IEEE Xplore