Automatic generation of statistical language models for interactive voice response applications

摘要: A Statistical Language Model (SLM) that can be used in an ASR for Interactive Voice Response (IVR) systems general and Natural Speech Applications (NLSAs) particular created by first manually producing a brief description text each task performed NLSA. These descriptions are then analyzed, one embodiment, to generate spontaneous speech utterances based pre-filler patterns skeletal set of content words. The turn with Part-of-Speech (POS) tagged conversations from corpus phrases. words is electronic lexico-semantic database thesaurus-based word extraction process more extensive list phrases set, thus generated, combined into using resource process. In statistical validation correct and/or add the automatically generated expected utterances. system requires minimum amount human intervention no prior knowledge regarding user utterances, WWW validate models. response prompt.