AUTOMATIC ALIGNMENT OF MAP TASK DIALOGS USING WFSTS

摘要: The goal of this work is the automatic alignment a map task dialog corpus collected for European Portuguese. Coral has been orthographically labeled, however off-the-shelf techniques do not because large amount cross-talk and pronunciation variation. This paper addresses these two issues. problem dealt with by using pre-processing stage channel separation, which proved specially advantageous in overlapping speaker turns. variation was addressed including alternative rules procedure. based on WFSTs sense that its search space defined distribution-to-word (or distributionto-phone) transducer. Despite many limitations, such as inadequacy our current acoustic phone models terms voice quality changes (such laughing), aligner sufficiently robust demonstrated feasibility implementation.