Speech databases used for training and adaptation comprise the stored audio of exemplary speech for model learning (training) and testing; a transcription of the spoken content may be given, together with labels for phenomena such as emotion, age, or personality. It is common wisdom in automatic speech processing that training data should be as close as possible to the data used for testing. For some applications, read, i.e., non-spontaneous, non-realistic speech data will do because the data used for testing will be read as well; examples are the screening of pathological or non-native speech or speaker verification for access control.