Reconnaissance de la parole guidée par des transcriptions approchées

In many cases, an approximated transcript can be associated to speech signal : movies subtitles, scenario and theatre, summaries and radio broadcast. These transcripts correspond rarely to the exact word utterances. The goal of this work is to use these information to improve the performance of an automatic speech recognition (ASR) system with integration of information resulting from the transcripts. In this paper we use the partial transcript in order both to adapt the language model and to rescore the ASR word hypothesis when the partial transcript matches the input signal. Multiple applications are possible : to help deaf people to follow a play with closed caption aligned to the voice signal (with respect to performer variations), to watch a movie in another language using aligned closed captions, to transcript in real time debates or meetings.