Spoken Book Alignment Using Wfsts

The framework of this paper is a national project known as IPSOM, whose main goal is to improve the access to digitally stored spoken books, used primarily by the visually impaired community, by providing tools for easily detecting and indexing units (words, sentences, topics). Simultaneously, the project also aims to broaden the usage of multimedia spoken books (for instance in didactic applications, etc.), by providing multimedia interfaces for access and retrieval. Hence, spoken book alignment is a major task. From the point of view of research, one of the most interesting aspects of the IPSOM project is the fact that indexed spoken books provide an invaluable resource for datadriven prosodic modelling and unit selection in the context of text-to-speech synthesis. This motivated doing the alignment not only on the basis of words, but rather sub-word units and also automatically generating multiple pronunciations by applying phonological rules in a WFST (Weighted Finite State Transducer) framework.