Discriminating languages in a probabilistic latent subspace

Aleksandr Sizov; Kong Aik Lee; Tomi Kinnunen

doi:10.21437/Odyssey.2016-12

Discriminating languages in a probabilistic latent subspace

Aleksandr Sizov, Kong Aik Lee, Tomi Kinnunen

Research output: Unpublished conference presentation (presented paper, abstract, poster) › Conference presentation (not published in journal/proceeding/book) › Academic research › peer-review

2 Citations (Scopus)

Abstract

We explore a method to boost discriminative capabilities of Probabilistic Linear Discriminant Analysis (PLDA) model without losing its generative advantages. To this end, our focus is in a low-dimensional PLDA latent subspace. We optimize the model with respect to MMI (Maximum Mutual Information) and our own objective functions, which is an approximation to the detection cost function. We evaluate the performance on NIST Language Recognition Evaluation 2015. Our model trains faster and performs more accurately in comparison to both generative PLDA and discriminative LDA baselines with 12% and 4% relative improvement in the average detection cost, respectively. The proposed method is applicable for a broad range of closed-set tasks.

Original language	English
Pages	81-88
Number of pages	8
DOIs	https://doi.org/10.21437/Odyssey.2016-12
Publication status	Published - Jun 2016
Externally published	Yes
Event	Speaker and Language Recognition Workshop, Odyssey 2016 - Bilbao, Spain Duration: 21 Jun 2016 → 24 Jun 2016

Conference

Conference	Speaker and Language Recognition Workshop, Odyssey 2016
Country/Territory	Spain
City	Bilbao
Period	21/06/16 → 24/06/16

ASJC Scopus subject areas

Signal Processing
Software
Human-Computer Interaction

More information

10.21437/Odyssey.2016-12

Cite this

@conference{2950a526a5d74ed98f458000d4548c66,

title = "Discriminating languages in a probabilistic latent subspace",

abstract = "We explore a method to boost discriminative capabilities of Probabilistic Linear Discriminant Analysis (PLDA) model without losing its generative advantages. To this end, our focus is in a low-dimensional PLDA latent subspace. We optimize the model with respect to MMI (Maximum Mutual Information) and our own objective functions, which is an approximation to the detection cost function. We evaluate the performance on NIST Language Recognition Evaluation 2015. Our model trains faster and performs more accurately in comparison to both generative PLDA and discriminative LDA baselines with 12% and 4% relative improvement in the average detection cost, respectively. The proposed method is applicable for a broad range of closed-set tasks.",

author = "Aleksandr Sizov and Lee, {Kong Aik} and Tomi Kinnunen",

note = "Publisher Copyright: {\textcopyright} Odyssey 2016: Speaker and Language Recognition Workshop. All rights reserved.; Speaker and Language Recognition Workshop, Odyssey 2016 ; Conference date: 21-06-2016 Through 24-06-2016",

year = "2016",

month = jun,

doi = "10.21437/Odyssey.2016-12",

language = "English",

pages = "81--88",

}

Discriminating languages in a probabilistic latent subspace. / Sizov, Aleksandr; Lee, Kong Aik; Kinnunen, Tomi.
2016. 81-88 Paper presented at Speaker and Language Recognition Workshop, Odyssey 2016, Bilbao, Spain.

Research output: Unpublished conference presentation (presented paper, abstract, poster) › Conference presentation (not published in journal/proceeding/book) › Academic research › peer-review

TY - CONF

T1 - Discriminating languages in a probabilistic latent subspace

AU - Sizov, Aleksandr

AU - Lee, Kong Aik

AU - Kinnunen, Tomi

PY - 2016/6

Y1 - 2016/6

N2 - We explore a method to boost discriminative capabilities of Probabilistic Linear Discriminant Analysis (PLDA) model without losing its generative advantages. To this end, our focus is in a low-dimensional PLDA latent subspace. We optimize the model with respect to MMI (Maximum Mutual Information) and our own objective functions, which is an approximation to the detection cost function. We evaluate the performance on NIST Language Recognition Evaluation 2015. Our model trains faster and performs more accurately in comparison to both generative PLDA and discriminative LDA baselines with 12% and 4% relative improvement in the average detection cost, respectively. The proposed method is applicable for a broad range of closed-set tasks.

AB - We explore a method to boost discriminative capabilities of Probabilistic Linear Discriminant Analysis (PLDA) model without losing its generative advantages. To this end, our focus is in a low-dimensional PLDA latent subspace. We optimize the model with respect to MMI (Maximum Mutual Information) and our own objective functions, which is an approximation to the detection cost function. We evaluate the performance on NIST Language Recognition Evaluation 2015. Our model trains faster and performs more accurately in comparison to both generative PLDA and discriminative LDA baselines with 12% and 4% relative improvement in the average detection cost, respectively. The proposed method is applicable for a broad range of closed-set tasks.

UR - http://www.scopus.com/inward/record.url?scp=85014917706&partnerID=8YFLogxK

U2 - 10.21437/Odyssey.2016-12

DO - 10.21437/Odyssey.2016-12

M3 - Conference presentation (not published in journal/proceeding/book)

AN - SCOPUS:85014917706

SP - 81

EP - 88

T2 - Speaker and Language Recognition Workshop, Odyssey 2016

Y2 - 21 June 2016 through 24 June 2016

ER -

Discriminating languages in a probabilistic latent subspace

Abstract

Conference

ASJC Scopus subject areas

More information

Other files and links

Fingerprint

Cite this