Experiments on the France telecom 3000 voice agency corpus
Géraldine Damnati, Frédéric Béchet, Renato De Mori · 2007
The recent advances in speech recognition technologies, and the experience acquired in the development of WEB or Interactive Voice Response interfaces, have facilitated the integration of speech modules in robust Spoken Dialog Systems (SDS), leading to the deployment on a large scale of speech-enabled services. With these services it is possible to obtain very large corpora of human-machine interactions by collecting system logs. This new kinds of systems and dialogue corpora offer new opportunities for academic research while raising two issues: How can academic research take profit of the system logs of deployed SDS in order to build the next generation of SDS, although the dialogues collected have a dialogue flow constrained by the previous SDS generation? On the other side, what immediate benefits can academic research offer for the improvement of deployed system? This paper addresses these aspects in the framework of the deployed France Telecom 3000 Voice Agency service.