1st Joint SIG-IL/Microsoft Workshop on Speech and Language Technologies for Iberian Languages
Porto Salvo, Portugal
Most of today's methods for transcription and indexation of
broadcast audio data are manual. Broadcasters process
thousands hours of audio and video data on a daily basis, in
order to transcribe that data, to extract semantic information,
and to interpret and summarize the content of those
documents. The development of automatic and efficient
support for these manual tasks has been a great challenge and
over the last decade there has been a growing interest in the
usage of automatic speech recognition as a tool to provide
automatic transcription and indexation of broadcast news and
random and relevant access to large broadcast news databases.
However, due to the common topic changing over time which
characterizes this kind of tasks, the appearance of new events
leads to high out-of-vocabulary (OOV) word rates and
consequently to degradation of recognition performance. This
is especially true for highly inflected languages like the
European Portuguese language.
Several innovative techniques can be exploited to reduce those errors. The use of news shows specific information, such as topic-based lexicons, pivot working script, and other sources such as the online written news daily available in the Internet can be added to the information sources employed by the automatic speech recognizer. In this thesis we are exploring the use of additional sources of information for vocabulary optimization and language model adaptation of a European Portuguese broadcast news transcription system. Hence, this thesis had 3 different main contributions: a novel approach for vocabulary selection using Part-Of-Speech (POS) tags to compensate for word usage differences across the various training corpora; language model adaptation frameworks performed on a daily basis for single-stage and multistage recognition approaches; a new method for inclusion of new words in the system vocabulary without the need of additional data or language model retraining.
Bibliographic reference. Martins, Ciro (2009): "Dynamic language modeling for european portuguese [phd thesis]", In SLTECH-2009, 113-114.