The presentation will focus on two major areas. First, approaches to a machine aided human translation (MAHT) problem where we attempt to improve automatic speech recognition performance for a human translator dictating translations in a target language while reading from a source language document are discussed. These approaches involve the integration of statistical acoustic and language models for ASR with statistical machine translation (SMT) models. Second, methods for developing efficient models of intrinsic speech variability are discussed. Approaches based on prior knowledge of articulatory dynamics and acoustic phonetics will be discussed along with data driven subspace based approaches.