Publication Details

Non-Negative Factor Analysis of Gaussian Mixture Model Weight Adaptation for Language and Dialect Recognition

BAHARI Mohamad H., DEHAK Najim, VAN hamme Hugo, BURGET Lukáš, ALI Ahmed M. and GLASS Jim. Non-Negative Factor Analysis of Gaussian Mixture Model Weight Adaptation for Language and Dialect Recognition. IEEE/ACM TRANSACTIONS ON AUDIO, SPEECH AND LANGUAGE PROCESSING, vol. 2014, no. 7, pp. 1117-1129. ISSN 2329-9290. Available from: http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=6803908
Czech title
Non-negativní faktorová analýza adaptace modelu se směsí Gaussovek pro rozpoznávání jazyka a dialektu
Type
journal article
Language
english
Authors
Bahari Mohamad H. (KUL)
Dehak Najim (JHU)
Van hamme Hugo (KUL)
Burget Lukáš, doc. Ing., Ph.D. (DCGM FIT BUT)
Ali Ahmed M. (QCRI)
Glass Jim (MIT)
URL
Keywords

Non-negative factor analysis; model adaptation; Gaussian mixture model weight; dialect recognition; language recognition

Abstract

In this research, a non-negative factor analysis approach is developed for GMM weight decomposition and adaptation. This methods allows for a new low-dimensional utterance representation similar to i-vectors.

Annotation

Recent studies show that Gaussian mixture model (GMM) weights carry less, yet complimentary, information to GMM means for language and dialect recognition. However, state-of-the-art language recognition systems usually do not use this information. In this research, a non-negative factor analysis (NFA) approach is developed for GMM weight decomposition and adaptation. This modeling, which is conceptually simple and computationally inexpensive, suggests a new low-dimensional utterance representation method using a factor analysis similar to that of the i-vector framework. The obtained subspace vectors are then applied in conjunction with i-vectors to the language/dialect recognition problem. The suggested approach is evaluated on the NIST 2011 and RATS language recognition evaluation (LRE) corpora and on the QCRI Arabic dialect recognition evaluation (DRE) corpus. The assessment results show that the proposed adaptation method yields more accurate recognition results compared to three conventional weight adaptation approaches, namely maximum likelihood re-estimation, non-negative matrix factorization, and a subspace multinomial model. Experimental results also show that the intermediate-level fusion of i-vectors and NFA subspace vectors improves the performance of the state-of-the-art i-vector framework especially for the case of short utterances.

Published
2014
Pages
1117-1129
Journal
IEEE/ACM TRANSACTIONS ON AUDIO, SPEECH AND LANGUAGE PROCESSING, vol. 2014, no. 7, ISSN 2329-9290
Publisher
IEEE Signal Processing Society
DOI
UT WoS
000338122000001
EID Scopus
BibTeX
@ARTICLE{FITPUB10731,
   author = "H. Mohamad Bahari and Najim Dehak and Hugo hamme Van and Luk\'{a}\v{s} Burget and M. Ahmed Ali and Jim Glass",
   title = "Non-Negative Factor Analysis of Gaussian Mixture Model Weight Adaptation for Language and Dialect Recognition",
   pages = "1117--1129",
   journal = "IEEE/ACM TRANSACTIONS ON AUDIO, SPEECH AND LANGUAGE PROCESSING",
   volume = 2014,
   number = 7,
   year = 2014,
   ISSN = "2329-9290",
   doi = "10.1109/TASLP.2014.2319159",
   language = "english",
   url = "https://www.fit.vut.cz/research/publication/10731"
}
Back to top