Ji Ming, Peter O'Boyle, Francis Jack Smith: An HMM with optimized segment-dependent observations for speech recognition. EUROSPEECH 1995: 1475-1478