Sometimes Ylva (standard) is used as the selected voice according to the TTS log, due to timeouts when querying the DNN CereProc server. @bertfrees has increased the timeout limits, so this is currently a testing issue for MTM.