Institutional Repository
Technical University of Crete
EN  |  EL



My Space

A sub-optimal viterbi-like search for linear dynamic models classification

Oikonomidis Dimitrios, Diakoloukas Vasilis, Digalakis Vasilis

Full record

Year 2007
Type of Item Conference Publication
Bibliographic Citation D. Oikonomidis, V. Diakoloukas and V. Digalakis, "A sub-optimal viterbi-like search for linear dynamic models classification," presented at 8th Annual Conference of the International Speech Communication Association, Antwerp, Belgium, 2007
Appears in Collections


This paper describes a Viterbi-like decoding algorithm appliedon segment-models based on linear dynamic systems (LDMs).LDMs are a promising acoustic modeling scheme which can alleviateseveral of the limitations of the popular Hidden MarkovModels (HMMs). There are several implementations of LDMsthat can be found in the literature. For our decoding experimentswe consider general identifiable forms of LDMs which allowincreased state space dimensionality and relax most of the constraintsfound in other approaches. Results on the AURORA2database show that our decoding scheme significantly outperformsstandard HMMs, particularly under significant noise levels.