Overcoming HMM time independence assumption using N-gram based modelling for continuous speech recognition

The development of new acoustical models that overcome traditional HMM restrictions is an active field of research in automatic speech recognition. One possible approach to achieve this goal is to work with N-gram based augmented HMM. In this paper, we propose to deal with time independence assumption of HMM using N-gram based modelling. For this, the temporal dependencies of each acoustic feature are explicitly modelled. Results obtained in this work testing this approach in continuous speech recognition show the suitability of adding long span information for ASR performance. Moreover, we are improving previous results obtained using this modelling scheme in connected digit recognition experiments.