benefitsofcondition-matchedcohortselectioninscorenormalization ... · 2017-08-28 · 0.1 0.2 0.5 1...
1
Analysis of mutual duration and noise effects in speaker recognition: benefits of condition-matched cohort selection in score normalization Andreas Nautsch ? Rahim Saeidi † Christian Rathgeb ? Christoph Busch ? ? da/sec – Biometrics and Internet Security Research Group, Hochschule Darmstadt, Germany † Department of Signal Processing and Acoustics, Aalto University, Finland andreas.nautsch@{cased|h-da}.de Motivation • Real-life conditions in commercial and forensic applications • Vast broadness of conditions e.g.: – Sample completeness (duration) – Ambient noise (AC / CROWD) • True sample condition remains unknown due to e.g., SNR estimations • Unified Audio Characteristics seem promising for condition-matching score- normalization Research Questions Q1: How extensive are mutual effects to the per- formance of a State-of-the-Art system? Q2: Does condition-informed cohort selection benefit from statistic approaches rather than from condition-matched cohorts? Q3: Do conditions affect basic i-vector proper- ties, i.e. mean i-vectors? Related Work Unified Audio Characteristics • Single multivariate Gaussian models in original (raw) i-vector space Λ i ∼N (μ i , Σ),i =1,..., 55 – Condition-dependent μ i – Shared full Σ by pooling • Condition quality vector (q-vector) ~ q as posterior probabilities for i-vector ~ w : ~ q (i)= P ( ~ w | Λ i ) ∑ 55 i=1 P ( ~ w | Λ i ) Conventional AS Score Normalization • Adaptive & Symmetric – Refs vs. probe-alike cohort set – Prbs vs. reference-alike cohort set – Computation of: μ ref ,σ ref ,μ prb ,σ prb • AS-normalized scores S AS by averaged symmetric zero-norm of score S : S AS = 1 2 S - μ ref σ ref + S - μ prb σ prb Cohort Selection Criterion • Symmetric Kullback-Leibler divergence: 1 2 55 X i=1 ~ q a (i) log ~ q a (i) ~ q b (i) + ~ q b (i) log ~ q b (i) ~ q a (i) • Selecting top-c cohort q-vectors Experimental Set-up & Results Condition 1 2 3 4 5 6 7 8 9 10 11 . . . 15 16 . . . 30 31 . . . 55 Duration 5s 10 s 20 s 40 s full 5s 10 s 20 s . . . full 5s ... full Noise clean Air Conditioner (AC) CROWD SNR 0 dB 5 dB 10 dB 15 dB 20 dB 0 ... 20 dB 0 dB . . . 20 dB 0 dB . . . 20 dB I4U Data of NIST SRE’12 • Condition-dependent sample versions from long-duration & clean samples • VAD labels of clean samples • LDA: 400 to 200 • PLDA: 200 factors • No calibration Baseline Results: impact of mutual conditions 5s 10 s 20 s 40 s full 0.01 0.1 0.2 0.5 1 2 5 Duration C llr 5s 10 s 20 s 40 s full 0.01 0.1 0.2 0.5 1 2 5 Duration C llr 0 dB 5 dB 10 dB 15 dB 20 dB Clean C llr C min llr (a) AC noise (b) CROWD noise Comparison of AS-norm approaches 0 5 10 15 20 0.1 0.3 0.5 0.7 SNR (in dB) C min llr 5 s 10 s 20 s 40 s full 0.1 0.3 0.5 0.7 Duration C min llr (c) 10s/CROWD (d) CROWD-0 dB no norm uAS (matched cohorts) cAS uAS (automatic cohort selection) Automatic cohort pre-selection 200 400 600 800 10 20 30 40 50 Cohort speaker ID Condition (e) Unique selections on 10s/CROWD-0 dB Cohort pre-selection by: • Similar conditions • SNR more relevant than duration Analysis: i-vector pool mean shift • Multi-variate Student t-test: t 2 = n x n y n x + n y ( ~ ¯ x - ~ ¯ y ) 0 W -1 ( ~ ¯ x - ~ ¯ y ), with: W = ∑ n x i=1 ( ~x i - ~ ¯ x)( ~x i - ~ ¯ x) 0 + ∑ n y i=1 (~ y i - ~ ¯ y )(~ y i - ~ ¯ y ) 0 n x + n y - 2 • P-values by cdf F of χ 2 (D = 200): t 2 ∼ χ 2 D , p =1 - F χ 2 D (t 2 ) 10 20 30 40 50 10 20 30 40 50 Condition Condition 0 5 e3 10 e3 15 e3 20 e3 (f) t 2 scores of Hotelling’s T 2 -test ⇒ High significance: p ( =1 same condition: same mean i-vector < 10 -13 cross condition: different mean i-vector Conclusion • Mutual effects with performance impacts by log-duration and log-SNR • Quality-based cohort pre-selection yields significant gains • Performance gap between degraded and non-degraded samples still as open chal- lenge • Means of cross-condition i-vectors differ Acknowledgements We would like to thank the I4U consortium for database sharing. This work has been partially funded by the Cen- ter for Advanced Security Research Darmstadt (CASED), the Hesse government (project no. 467/15-09, BioMobile) and the Academy of Finland (project no. 256961 and 284671).
Transcript of benefitsofcondition-matchedcohortselectioninscorenormalization ... · 2017-08-28 · 0.1 0.2 0.5 1...
Analysis ofmutual durationandnoise effects in speaker recognition:benefits of condition-matchedcohort selection in scorenormalization
Andreas Nautsch? Rahim Saeidi†
Christian Rathgeb? Christoph Busch?
? da/sec – Biometrics and Internet Security Research Group, Hochschule Darmstadt, Germany† Department of Signal Processing and Acoustics, Aalto University, Finland
andreas.nautsch@{cased|h-da}.de
Motivation
• Real-life conditions in commercial andforensic applications
• Vast broadness of conditions e.g.:
– Sample completeness (duration)– Ambient noise (AC / CROWD)
• True sample condition remains unknowndue to e.g., SNR estimations
• Unified Audio Characteristics seempromising for condition-matching score-normalization
Research Questions
Q1: How extensive are mutual effects to the per-formance of a State-of-the-Art system?
Q2: Does condition-informed cohort selectionbenefit from statistic approaches ratherthan from condition-matched cohorts?
Q3: Do conditions affect basic i-vector proper-ties, i.e. mean i-vectors?
Related Work
Unified Audio Characteristics
• Single multivariate Gaussian models inoriginal (raw) i-vector spaceΛi ∼ N (µi,Σ), i = 1, . . . , 55
– Condition-dependent µi– Shared full Σ by pooling
• Condition quality vector (q-vector) ~q asposterior probabilities for i-vector ~w:
~q(i) =P (~w |Λi)∑55i=1 P (~w |Λi)
Conventional AS Score Normalization
• Adaptive & Symmetric
– Refs vs. probe-alike cohort set
– Prbs vs. reference-alike cohort set
– Computation of: µref, σref, µprb, σprb
• AS-normalized scores SAS by averagedsymmetric zero-norm of score S:
SAS =1
2
(S − µref
σref+S − µprb
σprb
)Cohort Selection Criterion
• Symmetric Kullback-Leibler divergence:
1
2
55∑i=1
~qa(i) log~qa(i)
~qb(i)+ ~qb(i) log
~qb(i)
~qa(i)
• Selecting top-c cohort q-vectors
Experimental Set-up & Results
Condition 1 2 3 4 5 6 7 8 9 10 11 . . . 15 16 . . . 30 31 . . . 55
Duration 5 s 10 s 20 s 40 s full 5 s 10 s 20 s . . . full 5 s . . . full
Noise clean Air Conditioner (AC) CROWDSNR 0 dB 5 dB 10 dB 15 dB 20 dB 0 . . . 20 dB 0 dB . . . 20 dB 0 dB . . . 20 dB
I4U Data of NIST SRE’12
• Condition-dependent sampleversions from long-duration &clean samples
• VAD labels of clean samples
• LDA: 400 to 200
• PLDA: 200 factors
• No calibration
Baseline Results: impact of mutual conditions
5 s 10 s 20 s 40 s full
0.01
0.10.2
0.512
5
Duration
Cllr
5 s 10 s 20 s 40 s full
0.01
0.10.2
0.512
5
Duration
Cllr
0 dB
5 dB
10 dB
15 dB
20 dB
Clean
Cllr
Cminllr
(a) AC noise (b) CROWD noise
Comparison of AS-norm approaches
0 5 10 15 20
0.1
0.3
0.5
0.7
SNR (in dB)
Cm
inllr
5 s 10 s 20 s 40 s full
0.1
0.3
0.5
0.7
Duration
Cm
inllr
(c) 10s/CROWD (d) CROWD-0 dB
no norm uAS (matched cohorts)cAS uAS (automatic cohort selection)
Automatic cohort pre-selection
200 400 600 800
10
20
30
40
50
Cohort speaker ID
Con
dition
(e) Unique selections on 10s/CROWD-0 dB
Cohort pre-selection by:
• Similar conditions
• SNR more relevant than duration
Analysis: i-vector pool mean shift
• Multi-variate Student t-test:
t2 =nx nynx + ny
(~̄x− ~̄y)′W−1(~̄x− ~̄y),with:
W =
∑nx
i=1 (~xi − ~̄x)(~xi − ~̄x)′ +∑ny
i=1 (~yi − ~̄y)(~yi − ~̄y)′
nx + ny − 2
• P-values by cdf F of χ2 (D = 200):
t2 ∼ χ2D, p = 1− Fχ2
D(t2)
10 20 30 40 50
10
20
30
40
50
Condition
Con
dition
0
5 e3
10 e3
15 e3
20 e3
(f) t2 scores of Hotelling’s T 2-test
⇒ High significance: p
{= 1 same condition: same mean i-vector< 10−13 cross condition: different mean i-vector
Conclusion
• Mutual effects with performance impactsby log-duration and log-SNR
• Quality-based cohort pre-selection yieldssignificant gains
• Performance gap between degraded andnon-degraded samples still as open chal-lenge
• Means of cross-condition i-vectors differ
Acknowledgements
We would like to thank the I4U consortium fordatabase sharing.
This work has been partially funded by the Cen-ter for Advanced Security Research Darmstadt(CASED), the Hesse government (project no.467/15-09, BioMobile) and the Academy ofFinland (project no. 256961 and 284671).
![Quantitative visualization of flow inside an automotive ...-0.2-0.1 0.0 0.0 0.1 0.2 0.3x [m] y [m] V [m/s] y xy x A a) Cool-vent mode V [m/s]-0.3-0.2-0.1 0.0 0.0 0.1 0.2 0.3x [m] y](https://static.fdocuments.in/doc/165x107/6005c0f59988167be95732aa/quantitative-visualization-of-flow-inside-an-automotive-02-01-00-00-01.jpg)
