REMAP - Recursive Estimation and Maximization of A Posteriori Probabilities

Although very general, the method is developed in the context of a statistical model for transition-based speech recognition using Artificial Neural Networks (ANN) to generate probabilities for Hidden Markov Models (HMMs). In the new approach, we use local conditional posterior probabilities of transitions to maximize global posterior probabilities of word sequences. Although we still use ANNs to estimate posterior probabilities, the network is trained with targets that are themselves estimates of local posterior probabilities that are recursively estimated to guarantee a monotonic increase of global posteriors up to a (local) maximum. Initial experimental results support the theory by showing an increase in the estimates of posterior probabilities of the correct sentences after REMAP iterations, and statistically significant decrease in error rates for independent test sets.

Remap is a result of ongoing collaboration between Yochai Konig, Hervé Bourlard, and Nelson Morgan, and is developed at the speech group that is a part of the Realization Group at the International Computer Science Institute.

Papers

• Konig. Y., `` REMAP: Recursive Estimation and Maximization of A Posteriori Probabilities - Application to Transition-Based Connectionist Speech Recognition.'', Ph.D thesis, University of California at Berkeley.
• Konig. Y., Bourlard. H., and Morgan. N.,`` REMAP: Experiments with Speech Recognition.'', In ICASSP'96, Atlanta, GA.
• Bourlard, H., Konig. Y., and Morgan, N., `` REMAP: Recursive Estimation and Maximization of A Posteriori Probabilities in Connectionist Speech Recognition.'' In Eurospeech 95, Madrid, Spain, September 1995.
• Konig. Y., Bourlard. H., and Morgan, N., `` REMAP Modeling for Connectionist Speech Recognition.'' In Speech Research Symposium XV, Johns Hopkins University, MD, June 1995.
• Konig. Y., Bourlard. H., and Morgan. N., ``REMAP: Recursive Estimation and Maximization of A Posteriori Probabilities, Application to Transition-Based Connectionist Speech Recognition.'' To Appear in NIPS'95, Denver, CO, 1995.
• Bourlard. H., Konig. Y., and Morgan, N., `` REMAP: Recursive Estimation and Maximization of A Posteriori Probabilities, Application to Transition-Based Connectionist Speech Recognition.'' ICSI Technical Report TR-94-064, Berkeley, CA, March 1995.