Title: Finite-state Transducers for Speech-input Translation

Abstract: Nowadays, hidden Markov models (HMMs) and n-grams are the basic components of the best speech recognition systems. Similar models can be used for speech translation. Basically, the HMMs can be integrated into a finite-state transducer in a similar way as the acoustic models (HMMs) are integrated into a language model (n-gram) for speech decoding. Moreover, the translation process can be performed by searching for an optimal path of states in the integrated network. The output of this search process is a target word sequence associated to optimal path. HMMs can be trained from a speech corpus, and the translation model can be learnt automatically from a parallel training corpus. This approach have been assessed in the framework of the EuTrans project, founded by the European Union. Extensive speech-input experiments have been carried out from with Spanish to English and from Italian to English translation, in an application involving the interaction (by telephone) of a customer with a receptionist at the front-desk of a hotel.

Curriculum: Francisco Casacuberta received the Licenciatura (Master) and Doctorado (Ph.D.) in Physics degrees from the Universidad de Valencia, Spain, in 1976 and 1981, respectively. From 1976 to 1979 he worked with the Department of Electricity and Electronics at the Universidad de Valencia as an FPI fellow. From 1980 to 1985 he was with the Computing Center of the Universidad de Valencia as a systems analyst. Since 1986, he has been with the Department of Information Systems and Computation of the Universidad Politécnica de Valencia first as a Profesor Titular (Associate Professor) and from 1990 as a Catedrático (Full Professor). Since 1981, he has been an active member of research groups in the field of automatic speech recognition and translation. His current research interest lies in the areas of automatic speech recognition and translation, pattern recognition and machine learning. Dr. Casacuberta is a member of the Spanish Society for Pattern Recognition and Image Analysis (AERFAI), which is an affiliate society of International Association of Pattern Recognition (IAPR), the IEEE Computer Society and the Spanish Association for Artificial Intelligence (AEPIA).

Casacube.ps Casacube.pdf