Irene Clemente, Martin Heckmann, Gerhard Sagerer, Frank Joublin, "Multiple Sequence Alignment Based Bootstrapping for Improved Incremental Word Learning", International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2010.

We investigate incremental word learning with few training exam- ples in a Hidden Markov Model (HMM) framework suitable for an interactive learning scenario with little prior knowledge. When using only a few training examples the initialization of the models is a cru- cial step. In the bootstrapping approach proposed, an unsupervised initialization of the parameters is performed, followed by the retrain- ing and construction of a new HMM using multiple sequence align- ment (MSA). Finally we analyze discriminative training techniques to increase the separability of the classes using minimum classifica- tion error (MCE). Recognition results are reported on isolated digits taken from the TIDIGITS database.