• Partner:

  • DFKI

  • TU Berlin




Within the SpeechEval project we were facing many challenges and implemented different functionalities. In the following we shortly list some of our main features and system charateristics:
  • Automated end-to-end tests without human callers: Realization of an end-to-end user simulation, based on probabilistic features extracted from a corpus.
  • Functionality testing: Testing of specific SDS sub-components, e.g. automatic speech recognition (ASR), dialog manager (DM), etc.
  • Scripted dialogs: Definition of a specific dialog flow for the verification of specific dialog situations.
  • Stay-alive testing: Short test to examine the accessibility of a system.
  • Domain and System independence: Our user simulation is based on a large human-machine corpus, which covers many different state-of-the-art systems covering many different domains. Additionally, our simulation approach is domain independent. For more information see our publications.
  • Fast development time: We are extracting our knowledge sources semi-automatically from the corpus. This reduces the development time for setting up a new simulation.