* Ensure transformed version of a model is the same duration as the original model (don't end it when the input data runs out, as was previously done, because this will result in too short a model when latency compensation is used)