Section: New Results

Temporal Coherency Criterion for Alignment Inference Algorithms

Participants : Philippe Cuvillier [PhD Student] , Arshia Cont.

The question of modeling time and duration is of utmost importance for stability and robustness of real-time alignment algorithms and constitute one of the major success factors for the Antescofo listening machine described in [2] . Meanwhile, regular algorithms undergo stability in highly uncertain environments where observations obtained from the signal are highly uninformative and temporal information is of crucial importance.

PhD student Philippe Cuvillier defined Coherency Criteria for such applications and attempted to formalize such criteria in terms of probabilistic models and inference algorithms in case of Hidden Semi-Markov Chains. The results show that not all probabilistic families meet such criteria including some commonly used by engineers and designers. Preliminary results are submitted for publications and experimental results are being pursued.