Abstract In speech recognition, independence assumptions are typically made to reduce the complexity of the training and recognition search problems. One of the more blatantly invalid assumptions is that acoustic observations of phonemes are generated[…]