Open AccessUnimodal late fusion for NIST i ‐vector challenge on speaker detection
Author(s) -
Ali Hazrat,
Garcez Artur S. d'Avila,
Tran Son N.,
Zhou Xianwei,
Iqbal Khalid
Publication year - 2014
Publication title -
electronics letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.375
H-Index - 146
eISSN - 1350-911X
pISSN - 0013-5194
DOI - 10.1049/el.2014.1207
Subject(s) - nist , speech recognition , computer science , fusion , speaker recognition , artificial intelligence , sensor fusion , pattern recognition (psychology) , linguistics , philosophy
Speaker detection is a very interesting machine learning task for which the latest i ‐vector challenge has been coordinated by the National Institute of Standards and Technology (NIST). A simple late fusion approach for the speaker detection task on the i ‐vector challenge is presented. The approach is based on the late fusion of scores from the cosine distance method (the baseline) and the scores obtained from linear discriminant analysis. The results show that by adapting the simple late fusion approach, the framework can outperform the baseline score for the decision cost function on the NIST i ‐vector machine learning challenge.