Development of Acoustic and Linguistic Resources for Research and Evaluation in Interactive Vocal Information Servers

摘要: This paper describes the setting up of a resource database for research and evaluation in domain interactive vocal information servers. All this development work took place project aiming at an advanced speech recognition system automatic processing telephone directory requests was performed on basis Swiss-French Polyphone (collected framework European SpeechDat project). Due to unavailability properly orthographically transcribed, consistently labeled tagged unconstrained (together with its associated lexicon) targeted area, we first concentrated annotation structuration spoken data order make it profitable lexical linguistic modeling results. A baseline then trained newly developed resources tested. Preliminary experiments showed relative improvement 46% Word Error Rate (WER) compared results previously obtained very similar but working unconsistent natural that originally available.