: Difference between revisions

From HLT@INESC-ID

No edit summary
 
Line 16: Line 16:


== Abstract ==
== Abstract ==
The principal aim of this project is to create a system automatic translation of broadcast news of Spanish for Portuguese. The accomplishment of this system needs the combination of two operations: The automatic speech recognition of the news in Spanish followed by the translation of this recognition obtained for the Portuguese. This work had several parts beginning for the study of the recognizer, with base in the system of recognition AUDIMUS.MEDIA , [Meinedo et al., 2003], developed in the center of research INESC-ID (Instituto de Engenharia de Sistemas de Computadores) of Lisbon (Portugal), acquiring and trying material of audio and a text, constructing a lexicon and creating new acoustic models and language models for the recognizer of Spanish. There was obtained a result of evaluation similar to other sistems of recognition international and that appear in the state of art. For the translator based on (Statistical Machine Translation), they are in use parallel normalized texts, for both languages in study, Spanish and Portuguese, there are trained the probabilities of translation of the system, and aspect a human not exhaustive evaluation, due to the period of execution of the project, and the lack of sufficient personnel for a deeper study, staying for future lines a more detailed and joint evaluation of the global system.


== Resumo ==
== Resumo ==

Revision as of 20:39, 3 November 2007

Raquel Sánchez
Raquel Sánchez

Date

  • 14:30, November 02, 2007
  • Room 336

Speaker

Abstract

The principal aim of this project is to create a system automatic translation of broadcast news of Spanish for Portuguese. The accomplishment of this system needs the combination of two operations: The automatic speech recognition of the news in Spanish followed by the translation of this recognition obtained for the Portuguese. This work had several parts beginning for the study of the recognizer, with base in the system of recognition AUDIMUS.MEDIA , [Meinedo et al., 2003], developed in the center of research INESC-ID (Instituto de Engenharia de Sistemas de Computadores) of Lisbon (Portugal), acquiring and trying material of audio and a text, constructing a lexicon and creating new acoustic models and language models for the recognizer of Spanish. There was obtained a result of evaluation similar to other sistems of recognition international and that appear in the state of art. For the translator based on (Statistical Machine Translation), they are in use parallel normalized texts, for both languages in study, Spanish and Portuguese, there are trained the probabilities of translation of the system, and aspect a human not exhaustive evaluation, due to the period of execution of the project, and the lack of sufficient personnel for a deeper study, staying for future lines a more detailed and joint evaluation of the global system.

Resumo

Retrieved from ""