System and method for building optimal state-dependent statistical utterance classifiers in spoken dialog systems

摘要: A system and a method to generate statistical utterance classifiers optimized for the individual states of spoken dialog is disclosed. The make use large databases transcribed annotated utterances from calls collected in production log data reporting association between state at moment when were recorded utterance. From state, being vector multiple variables, subsets these certain variable ranges, quantized values, etc. can be extracted produce multitude distinct matching every possible state. For each subset combinations, trained, tuned, tested, stored together with performance results combination. Once set have been put into system, given resulting optimum selected result list used perform classification.