Main website: http://opendial-toolkit.net.
OpenDial is a Java-based, domain-independent toolkit for developing spoken dialogue systems. The primary focus of OpenDial is on robust and adaptive dialogue management, but OpenDial can also be used to build full-fledged, end-to-end dialogue systems, integrating speech recognition, language understanding, generation and speech synthesis.
The purpose of OpenDial is to combine the benefits of logical and statistical approaches to dialogue modelling into a single framework. The toolkit relies on probabilistic rules to represent the domain models in a compact and human-readable format. Supervised or reinforcement learning techniques can be applied to automatically estimate unknown rule parameters from relatively small amounts of data (see Lison (2014) for details). The hybrid approach adopted by OpenDial makes it possible to incorporate expert knowledge and domain-specific constraints in a robust, probabilistic framework.
OpenDial is designed as a blackboard architecture in which all system modules are connected to a central information hub representing the dialogue state (which is encoded as a Bayesian Network). Modules can therefore be plugged in and out of the system without affecting the rest of the processing pipeline. A collection of plugins is available to connect external components (for speech recognition, parsing, speech synthesis, etc.). New modules can also be easily implemented and integrated into the OpenDial architecture.
The toolkit has been originally developed by the Language Technology Group of the University of Oslo (Norway), with Pierre Lison as main developer.