REMORA A software architecture for the collaboration of different knowledge sources in phonetic decoding of continuous speech

REMORA, (REpresentation et Modelisation Objet pour la Reconnaissance Acoustico-phonetique: object representation and modelling for the acoustic phonetic recognition) is a software architecture for helping the phonetic decoding of continuous speech, using artificial intelligence techniques. Acoustic phonetic decoding is a central problem in continuous speech recognition. The basic idea of this project, realized at ENST, is to include various kinds of knowledge to improve accuracy and efficiency. The procedural knowledge may come from several sources such as centisecond segment recognition, diphone spotting or segmentation. The declarative knowledge may come from expertise on speech signal processing or from phonetic knowledge. In order to connect these two types of knowledge and to satisfy modularity and efficiency, we use an object-oriented formalism. The programming environment is a set of interactive menus, designed to be easily employed by the experts.