Incorporating phonetic knowledge into a multi-stream HMM framework | IEEE Conference Publication | IEEE Xplore