Sorry, you need to enable JavaScript to visit this website.

facebooktwittermailshare

FACTORIZED HIDDEN VARIABILITY LEARNING FOR ADAPTATION OF SHORT DURATION LANGUAGE IDENTIFICATION MODELS

Abstract: 

Bidirectional long short term memory (BLSTM) recurrent neural networks (RNNs) have recently outperformed other state-of-the-art approaches, such as i-vector and deep neural networks (DNNs) in automatic language identification (LID), particularly when testing with very short utterances (∼3s). Mismatches conditions between training and test data, e.g. speaker, channel, duration and environmental noise, are a major source of performance degradation for LID. A factorized hidden variability subspace (FHVS) learning technique is proposed for the adaptation of BLSTM RNNs to compensate for these types of mismatches in recording conditions. In the proposed approach, condition dependent parameters are estimated to adapt the hidden layer weights of the BLSTM in the FHVS. We evaluate FHVS on the AP17-OLR data set. Experimental results show that the FHVS method outperforms the standard BLSTM approach, achieving 27% relative improvements with utterance-level adaptation over the standard BLSTM for 1s duration utterances.

POSTER.pdf

PDF icon POSTER.pdf (698 downloads)
up
0 users have voted:

Paper Details

Authors:
Sarith Fernando, Vidhyasaharan Sethu, Eliathamby Ambikairajah
Submitted On:
12 April 2018 - 9:48pm
Short Link:
Type:
Poster
Event:
Presenter's Name:
Sarith Fernando
Paper Code:
SP-P4.7
Document Year:
2018
Cite

Document Files

POSTER.pdf

(698 downloads)

Subscribe

[1] Sarith Fernando, Vidhyasaharan Sethu, Eliathamby Ambikairajah, "FACTORIZED HIDDEN VARIABILITY LEARNING FOR ADAPTATION OF SHORT DURATION LANGUAGE IDENTIFICATION MODELS", IEEE SigPort, 2018. [Online]. Available: http://sigport.org/2551. Accessed: Apr. 21, 2018.
@article{2551-18,
url = {http://sigport.org/2551},
author = {Sarith Fernando; Vidhyasaharan Sethu; Eliathamby Ambikairajah },
publisher = {IEEE SigPort},
title = {FACTORIZED HIDDEN VARIABILITY LEARNING FOR ADAPTATION OF SHORT DURATION LANGUAGE IDENTIFICATION MODELS},
year = {2018} }
TY - EJOUR
T1 - FACTORIZED HIDDEN VARIABILITY LEARNING FOR ADAPTATION OF SHORT DURATION LANGUAGE IDENTIFICATION MODELS
AU - Sarith Fernando; Vidhyasaharan Sethu; Eliathamby Ambikairajah
PY - 2018
PB - IEEE SigPort
UR - http://sigport.org/2551
ER -
Sarith Fernando, Vidhyasaharan Sethu, Eliathamby Ambikairajah. (2018). FACTORIZED HIDDEN VARIABILITY LEARNING FOR ADAPTATION OF SHORT DURATION LANGUAGE IDENTIFICATION MODELS. IEEE SigPort. http://sigport.org/2551
Sarith Fernando, Vidhyasaharan Sethu, Eliathamby Ambikairajah, 2018. FACTORIZED HIDDEN VARIABILITY LEARNING FOR ADAPTATION OF SHORT DURATION LANGUAGE IDENTIFICATION MODELS. Available at: http://sigport.org/2551.
Sarith Fernando, Vidhyasaharan Sethu, Eliathamby Ambikairajah. (2018). "FACTORIZED HIDDEN VARIABILITY LEARNING FOR ADAPTATION OF SHORT DURATION LANGUAGE IDENTIFICATION MODELS." Web.
1. Sarith Fernando, Vidhyasaharan Sethu, Eliathamby Ambikairajah. FACTORIZED HIDDEN VARIABILITY LEARNING FOR ADAPTATION OF SHORT DURATION LANGUAGE IDENTIFICATION MODELS [Internet]. IEEE SigPort; 2018. Available from : http://sigport.org/2551