• Home
  • /All news
  • /The latest additions in RAPP’s platform services
The latest additions in RAPP’s platform services

The latest additions in RAPP’s platform services

RAPP is incorporating a custom ontology that will promote collective knowledge and cooperation between the participating robots along with a state of the art speech recognition service.

Ontology

We are combining the Knowrob Robot Ontology with the openAAL ambient assisted living ontology in order to create a custom ontology that will store and share the collectively acquired knowledge of all robots participating in RAPP. Robots will contribute to the collective knowledge by uploading information, patterns and behaviors that match certain already known or externally (by the User) asserted characteristics. When a robot is reluctant to proceed with an action it can consult the ontology and learn from the past experiences of other robots. This is a powerful tool that promotes cooperation between robots and significantly increases their total learning rate and knowledge capacity. In time, the collective knowledge will expand and new robots will successfully cope with unusual situations and behaviors they never encountered before and all that with absolutely no user intervention!

ontology

Speech recognition

A speech recognition toolkit will be a key interaction component between the user and microphone equipped robots (like NAO). It must be able to successfully deal with different use case scenarios and also understand all kinds of speakers coming from different age groups and nationalities. For this reason we decided to use CMUSphinx, a state of the art speech recognition toolkit developed at Carnegie Mellon University. CMUSphinx is a powerful and feature rich toolkit that provides the flexibility needed to deal with situations where the requirements are contradicting like a dictation scenario where a low number of mismatches is required and a command and control application where the key is speed and accuracy. Some of its most prominent features we will be using are acoustic adaptation, meaning the ability to adapt to a specific speaker and thus better recognize his voice, use of a grammar model to strictly define command and control applications and its training algorithms should we decide to expand its detection abilities to recognize new words and languages.

This article was written by Athanassios Kintsakis, PhD Student, Department of Electrical and Computer Engineering, Aristotle University of Thessaloniki (AUTH)