Vystadial 2016 is a dataset of telephone conversations in Czech, developed for training acoustic models for automatic speech recognition in spoken dialogue systems. The data comprise over over 75 hours in Czech, plus orthographic transcriptions.