Research Focus: Spoken Dialog Systems

Spoken dialog systems, aka Interactive Voice Response (IVR) systems, can render services on the phone and other speech input/output channels similar to (and beyond) what human assistants can do.  A recent example is the voice assistant Siri which is able to operate many applications on the iPhone as well as respond to a vast number of user requests including e-mail dictation, question answering, and search.  Since the early 2000s, spoken dialog systems have been widely deployed in customer contact centers (aka call centers) to automate services such as

  • directory services
  • call routing
  • phone banking
  • automatic troubleshooting for digital service providers
  • package, flight, and order tracking
High-Level Architecture of a Spoken Dialog System
Example Software Stack of an Open-Source Spoken Dialog System

(click the pictures to enlarge)

The research focus Spoken Dialog Systems at the DHBW Stuttgart maintains tight relationships with German and international academic institutions (e.g. Carnegie Mellon University, Pittsburgh and Ulm University) as well as industrial partners (e.g. Synchronoss, Bridgewater and ICSI, Berkeley).  Well-equipped in terms of computational resources (a Linux cluster with about 100 virtual machines, well over 1TB of RAM, and dozens of terabytes of hard drive space), the research focus makes primarily use of open-source components to perform cutting-edge R&D in the domain of spoken language understanding, dialog management, speech recognition, machine learning, among others, as proven by a substantial number of peer-reviewed publications in books, journals, and at international conferences as well as invited talks and presentations (e.g., at NAACL in Montreal, Interspeech in Florence, or SpeechTEK in New York).  Flagship project of the research focus is the spoken dialog system Halef .


  • HALEF: Help Assistant—Language-Enabled and Free
  • OASIS: Open-Source Automatic Speech Recognition In Smart Devices


Prof. Dr.-Ing. David Sündermann /