Ahilan's blog


ASSTA new researcher award

posted Oct 10, 2014, 12:32 PM by Ahilan M.K   [ updated Oct 10, 2014, 12:33 PM ]

Recently, I was awarded ASSTA new researcher award by Speech Science and Technology (SST) conference committee. The award consists of:
a) AU$750 grant
b) cost of the conference dinner
c) presentation of certificate at the conference dinner

Recently submitted ICASSP paper - Domain Adaptation for PLDA based Speaker Verification with Limited In-domain Data

posted Oct 10, 2014, 12:31 PM by Ahilan M.K   [ updated Jan 14, 2015, 3:42 PM ]

My co-researcher studied the probabilistic linear discriminant analysis (PLDA) speaker verification approach with limited in-domain (NIST SREs) data. Speaker verification performance is investigated when a limited number of speakers and sessions/speaker data are available in target in-domain (NIST SREs) and a much larger amount of data are available in out-domain (SWB). Subsequently, a number of techniques, including linear-weighted and pooled approaches are investigated to improve the speaker verification performance. There research outcomes were submitted to ICASSP 2015 conference.

Recently submitted ICASSP paper - The QUT-NOISE-SRE Protocol for the Evaluation of Noisy Speaker Recognition

posted Oct 10, 2014, 12:31 PM by Ahilan M.K

My co-researcher designed the QUT-NOISE-SRE protocol to mix the previously-created QUT-NOISE database, consisting of over 10 hours of background noise, collected across 10 unique locations covering 5 common noise scenarios, with commonly used speaker recognition datasets such as Switchboard, Mixer and the speaker recognition evaluation (SRE) datasets provided by NIST. In this work, we use the QUT-NOISE-SRE protocol to evaluate a state-of-the-art PLDA i-vector speaker recognition system. There research outcomes were submitted ICASSP 2015 conference.

Recently submitted ICASSP paper - Improving Short Utterance PLDA Speaker Verification using Multiple Enrolment Utterances

posted Oct 10, 2014, 12:30 PM by Ahilan M.K

Our recent studies have found that instead of using single long-utterance as enrolment data, if long enrolled-utterance is segmented into multiple short utterances and average of short utterance i-vectors is used as enrolled data, that improves the Gaussian PLDA (GPLDA) speaker verification. Subsequently, short utterance variance (SUV)-PLDA is also studied with multiple enrolment utterances. These research outcomes were submitted to ICASSP 2015 conference.

Recently submitted ICASSP paper - Weighted Median Fisher Discriminator and Linear-weighted Approaches to Limited Development Data PLDA Speaker Verification

posted Oct 10, 2014, 12:28 PM by Ahilan M.K

Recently, we studied the PLDA speaker verification system with limited microphone data, and we proposed the addition of a weighted median Fisher discriminator (WMFD) projection prior to length-normalised Gaussian probabilistic linear discriminant analysis (GPLDA) modelling. In combination with a linear-weighted approach to calculating cross-channel GPLDA parameters, WMFD-projected GPLDA extracts more reliable speaker discriminant information from limited development data, and provides a improvement over standard LDA-projected length-normalised GPLDA in interview-interview condition. These research outcomes were submitted to ICASSP 2015 conference.

Recently submitted ICASSP paper - Improving PLDA Speaker Verification using Utterance Partitioning Approach in Limited Development Data Conditions

posted Oct 10, 2014, 12:27 PM by Ahilan M.K

In recent studies, we have found that when speaker verification is evaluated on 10sec-10sec condition, at least around 40sec speech utterances are required for PLDA modelling. We also introduce an utterance partitioning approach in PLDA speaker verification which shows improvement over the baseline approach in limited session data conditions. These research outcomes were submitted to ICASSP 2015 conference.

Recently submitted ICASSP paper - Improving Out-domain PLDA Speaker Verification using Unsupervised Inter-dataset Variability Compensation Approach

posted Oct 10, 2014, 12:25 PM by Ahilan M.K

Recently researchers have found that that when the state-of-the-art probabilistic linear discriminant analysis (PLDA) speaker verification systems are trained using out-domain data, it significantly affects speaker verification performance due to the mismatch between development data and evaluation data. To overcome this problem we propose a novel unsupervised inter dataset variability (IDV) compensation approach to compensate the dataset mismatch. This research outcome was submitted to ICASSP 2015 conference.

My publications

posted Oct 10, 2014, 12:23 PM by Ahilan M.K

List of my publications details, including published and submitted can be found in this link: http://www.ahilan83.com/publications

1-8 of 8