ATVS-UAM NIST LRE 2009 System Description
Entity
UAM. Departamento de Ingeniería InformáticaPublisher
National Institute of Standards and TechnologyDate
2009Citation
NIST Language Recognition Evaluation. National Institute of Standards and Technology, 2009Subjects
InformáticaNote
Official contribution of the National Institute of Standards and Technology; not subject to copyright in the United States.Abstract
ATVS-UAM submits a fast, light and efficient single system. The use of a task-adapted nonspeech-recognition-based VAD (apart from NIST conversation labels) and gender-dependent total variability compensation technology allows our submitted system to obtain excellent development results with SRE08 data with exceptional computational efficiency. In order to test the VAD influence in the evaluation results, a contrastive equivalent system has been submitted exclusively changing ATVS VAD labels with BUT publicly contributed ones. In all contributed systems, two gender-independent calibrations have been trained with respectively telephone-only and mic (either mic-tel, tel-mic or mic-mic) data. The submitted systems have been designed for English speech in an application-independent way, all results being interpretable in the form of
calibrated likelihood ratios to be properly evaluated with Cllr. Sample development results with English SRE08 data are 0.53% (male) and 1.11% (female) EER in tel-tel data (optimistic as all English speakers in SRE08 are included in total variability matrices), going up to 3.5% (tel-tel) to 5.1% EER (tel-mic) in pessimistic cross-validation experiments (25% of test speakers totally excluded from development data in each xval set). The submitted system is extremely light in computational resources, running 77 times faster than real time. Moreover, once VAD and feature extraction are performed (the heaviest components of our system), training and testing are performed respectively at 5300 and 2950 times faster than real time.
Files in this item
Google Scholar:López-Moreno, Ignacio
-
González-Domínguez, Joaquín
-
Franco-Pedroso, Javier
-
Ramos Castro, Daniel
-
Toledano, Doroteo T.
-
González Rodríguez, Joaquín
This item appears in the following Collection(s)
Related items
Showing items related by title, author, creator and subject.