Voice-driven services over the telephone networks in Europe are continuously increasing. Today, in such traditional voice-driven servers, the recogniser is embedded in the telecommunication network and test results show that mobile network low bit rate coding and transmission errors degrade the performance of speech recognition severely. As a solution, the Distributed Speech Recognition architecture (DSR) has been designed by the ETSI-AURORA Group in order to improve robustness of voice-driven servers against these networks effects. Indeed such Services for customers are currently emerging in mobile networks, and, two Operator Partners, Sonofon and Bouygues Telecom, are highly interested in a real context DSR validation, which is the major goal of TriaDis. Three other Partners - BOSCH for the Terminal, L&H and CPK for DSR - are offering their background to implement the DSR terminal-server platform for these Trials.
The main goal of TriaDis is to validate, through Trials with selected group of end-users, the "DSR" improvement in terms of performance and usability on Voice-controlled server embedded in a real Mobile Network (GSM, DCS-1800). DSR concept has been recently introduced, designed and evaluated by the ETSI-AURORA Group with the target to make a standard. Hence TriaDis plans to implement "DSR" on a real server application accessed by any mobile terminal and perform validation trials. Indeed no real-context trials have been done but only tests on simulated environments in the frame of ETSI-AURORA. Moreover, the Consortium possess an important background allowing to get ready in a very short-term a validation platform after adaptation from an existing Terminal prototype platform and speech-controlled dialogue application (L&H, CPK). If successful, high exploitation of TriaDis results is expected by the Consortium, namely by Operator Partners.
The project is divided into 4 workpackages, whose 2 are devoted to the specification/implementation, one is focused on Trials and last on Exploitation/Standard following:
a) to specify (WP1) and implement (WP2) the final version available of the Distributed Speech Recognition (DSR designed by ETSI-AURORA) concept on a terminal and voice-controlled server application embedded in two real Mobile Networks (GSM, DCS-1800). A specification of the terminal handset and the remote Server application (into Network) will also be done beforehand in WP1.
b) to validate (WP3) through the two types of real Mobile Networks, which has not been made so far (only simulations are available), the DSR implemented on the terminal and the remote server located in the mobile networks. This validation will be performed with a limited but representative enough group of end-users calling the remote server with their mobile terminals. The tests to be performed are based on complete services.
c) in WP4, to provide major contributions to ETSI on the use of DSR concept in real application when Trials results will be available and fully analysed. Still in the WP4 frame, these Trials on a real platform will allow to prepare the implementation of such voice-controlled services for the Operators and might provide, beyond the scope of this project, a first platform for the study of "Next-generation-DSR"-based voice-operated services to be adapted for future emerging types of networks (UMTS): such future application will be very likely based on voice-request/multimedia response.
Five milestones are defined to follow TriaDis progress:
a) T01 and T03: services (T01) and Network-embedded Server / Terminal Application based on DSR architecture (T03) specified
b) T09: Testbed platform implemented (mobile terminal & server application)
c) T11: Preliminary results about Trials of Testbed platform onto GSM & DCS-1800 networks
d) T15: End of Trials and Results, Exploitation by Operators & other Industrial Partners
Funding SchemeACM - Preparatory, accompanying and support measures
92100 Boulogne Billancourt