TC-STAR 2006 Automatic Speech Recognition Evaluation: The UVIGO System

This paper describes the ongoing development of the University of Vigo’s Automatic Speech Recognition system (UVIGO) for the automatic transcription of Spanish European Parliamentary Plenary sessions and Spanish Parliamentary sessions. The system was developed to partake in the 2006 TC-STAR Automatic Speech Recognition evaluation campaign in the Spanish language section. The UVIGO system was derived from the University of Vigo’s Galician Broadcast N ews (BN) Transcription system by adapting the BN acoustic and language models to the TC-STAR domain. A detailed discussion of the front-end processing, acoustic and language modelling, and decoding process are presented. The proposed decoding strategy was developed to make the best possible use of gender- and speakerdependent acoustic models without a prior gender or speaker identifica tion process. In addition to describing the system architecture and reporting the evaluation results, we also highlight further improvements tha t we are planning to make to the overall system.