Integration of acoustic, phonetic, prosodic and lexical knowledge in an expert system for speech understanding