A SEGMENT-BASED APPROACH TO SPEECH RECOGNITION USING ACOUSTIC OBSERVATION CONTEXT