Statistical language models for query-by-example spoken document retrieval