PoSTPort (POrting Speech Technologies to other varieties of Portuguese): Difference between revisions

From HLT@INESC-ID

No edit summary
Line 25: Line 25:
* [[Ernesto Pardal]]  (CLUL)
* [[Ernesto Pardal]]  (CLUL)
* [[Helena Moniz]] (INESC-ID/CLUL)
* [[Helena Moniz]] (INESC-ID/CLUL)
* [[Nelson Neto]] (Scholarship)
* [[Puneet Sahani]] (Scholarship)


Undergraduate Students:  
Undergraduate Students:  

Revision as of 18:29, 23 January 2009

Sponsored by: FCT (PTDC/PLP/72404/2006)
Start: January 2008
End: December 2010

News

Open Research Position in project PoSTPort - Porting Speech Technologies to other Varieties of Portuguese (2008)

Team

Project Leader: Isabel Trancoso

Undergraduate Students:

Summary

The goal of this project is porting spoken language technologies originally developed for European Portuguese to other varieties of Portuguese, namely those spoken in South-American and African countries. The two main spoken language technologies that will be investigated are text-to-speech and speech-to-text conversion, commonly known as synthesis and recognition. Instead of porting complete systems, we shall concentrate on the linguistically relevant modules, that are likely to be most affected. The main task of this proposal, therefore, is structured in terms of modules which will be ported and evaluated independently as much as possible. Prior to this main work, however, the project will involve two tasks. In the first one, we shall build the pilot corpus necessary for studying the main linguistic differences and porting to the new varieties. In the second task, we shall characterize the main differences between the studied varieties, starting with the orthographic and syntactic levels, and ending with the phonetic and phonological levels. As an outcome of this task, we shall attempt to define a phonetic alphabet that encompasses all varieties of Portuguese. The fourth task concerns the automatic identification of spoken varieties of Portuguese. This is an important goal by itself and also relevant as a pre-processing stage for switching among recognition systems developed for specific varieties. We plan to implement this identification module by exploring different types of cues, such as phonotactic, acoustic and prosodic ones. The four tasks of the proposal combine linguistic, signal processing and natural language processing knowledge, thus reflecting the importance of an interdisciplinary team.

Workplan

  • T1 - Pilot Corpus collection
  • T2 - Characterization of main differences
  • T3 - Porting spoken language modules
  • T4 - Automatic identification of spoken varieties
  • T5 - Management

Corpus

Publications

Jean-Luc Rouas, Isabel Trancoso, Maria do Céu Guerreiro Viana Ribeiro, Mónica Abreu, Language and variety verification on broadcast news for Portuguese, Speech Communication, Elsevier, vol. 50, n. 11, November 2008

Alberto Abad, Isabel Trancoso, The L2F language verification systems for Albayzin-08 evaluation, In V Jornadas en Tecnología del Habla, November 2008

Patrick Silva, Nelson Neto, Aldebaro Klautau, André Adami, Isabel Trancoso, Speech Recognition for Brazilian Portuguese using the Spoltech and OGI-22 Corpora, In XXVI SIMPOSIO BRASILEIRO DE TELECOMUNICACOES, Rio de Janeiro, Brazil, September 2008

Jean-Luc Rouas, Isabel Trancoso, Maria do Céu Guerreiro Viana Ribeiro, Mónica Abreu, Portuguese Variety Identification on Broadcast News, In ICASSP 2008 - Int. Conf. on Acoustics, Speech, and Signal Processing, IEEE, Las Vegas, USA, April 2008

Demos

This demo shows our current Broadcast News recognizer for Brazilian Portuguese.