SNR-lnvariant PLDA Modeling forRobust Speaker VerificationNaLiandMan-WaiMakInterspeech2015Dresden,GermanyDepartment of Electronic and Information EngineeringThe Hong Kong Polytechnic University, Hong Kong SAR, China
SNR-Invariant PLDA Modeling for Robust Speaker Verification Na Li and Man-Wai Mak Department of Electronic and Information Engineering The Hong Kong Polytechnic University, Hong Kong SAR, China Interspeech 2015 Dresden, Germany
ContentsBackground and Motivation of Work1. E2. SNR-invariant PLDA modeling for Robust SpeakerVerification3.ExperimentsonSRE124.Conclusions
Contents 1. Background and Motivation of Work 2. SNR-invariant PLDA modeling for Robust Speaker Verification 3. Experiments on SRE12 4. Conclusions
BackgroundI-vector/PLDA FrameworkX, = m+ Vh, +&jm : Global mean of all i-vectorsV : Bases of speaker subspaceh, : Latent speaker factor with a standard normaldistributiol N (O, I)&.. : Residual term follows a Gaussian distributionwith zero mean and full covariance N(o, 2)x. : Length-normalized i-vector of speaker i
Background • I-vector/PLDA Framework m : Global mean of all i-vectors V : Bases of speaker subspace : Latent speaker factor with a standard normal distribution : Residual term follows a Gaussian distribution with zero mean and full covariance : Length-normalized i-vector of speaker i x m Vh ij i ij = + + ε hi ij ε xij
BackgroundIn conventional multi-condition training, wepooli-vectors from various background noiselevels to train m, V and ZSNR1I-vectorswith2SNRrangesSNR2SpeakerEM→(m,V,E)AlgorithmSession30102040SNR/dB
Background • In conventional multi-condition training, we pool i-vectors from various background noise levels to train m, V and Σ. EM Algorithm {m,V,S} I-vectors with 2 SNR ranges
MotivationWe arguethat the variation caused by SNR can be modeledby anSNRsubspaceand utterances falling within a narrowSNRrange shouldsharethesamesetofSNRfactors.SNRSNRGroup1SubspaceFactor1十年SNRFactor2Group2SNRFactor3Group3!
Motivation • We argue that the variation caused by SNR can be modeled by an SNR subspace and utterances falling within a narrow SNR range should share the same set of SNR factors. SNR Subspace SNR Factor 2 Group1 Group2 Group3 SNR Factor 1 SNR Factor 3