Robustness of Quality-based Score Calibration of Speaker Recognition - - PowerPoint PPT Presentation

Robustness of Quality-based Score Calibration of Speaker Recognition Systems

with respect to low-SNR and short-duration conditions

Andreas Nautsch⋆, Rahim Saeidi†, Christian Rathgeb⋆, Christoph Busch⋆

⋆Hochschule Darmstadt, CRISP, CASED, da/sec Security Research Group † Department of Signal Processing and Acoustics, Aalto University, Finland

Odyssey’16, Bilbao, 24.06.2016

Nautsch, Saeidi, Rathgeb, Busch Quality-based Score Calibration / Odyssey’16, Bilbao, 24.06.2016 1/3

Motivation Quality-based Calibration and low-Quality Robustness

Calibration schemes in unconstrained conditions

◮ Conventional: train on full/clean ⇒ use on all

df: 2

◮ Matched: train and use for each condition

df: 2 C

◮ Quality Measure Function: duration/SNR/. . .

df: 3 in linear fusion: S′ = w0 + w1 S + w2 Q (λref, λprb) Unified Audio Characteristics

◮ Gaussian models for conditions on raw i-vector domain

as vector representation of quality estimates q

◮ Bi-linear combination matrix: w2

qref Σ qprb df: C2 + 3

◮ Proposed: Function of Quality Estimates (FQEs)

df: 3 e.g., by cosine distance: S′ = w0 + w1 S + w2 cos ( qref, qprb)

Nautsch, Saeidi, Rathgeb, Busch Quality-based Score Calibration / Odyssey’16, Bilbao, 24.06.2016 2/3

Results Robustness Analyses in 55 combined conditions

Matched Q1 Q2 Q1+2 Qqvec Conventional f u l l / 5 d B f u l l / d B 4 s / 5 d B 4 s / d B 2 s / 5 d B 2 s / d B 1 s / c l e a n 1 s / 2 d B 1 s / 1 5 d B 1 s / 1 d B 5 s / c l e a n 5 s / 2 d B 5 s / 1 5 d B 5 s / 1 d B 50 100 Calibration Condition Cmc ×103

low SNR short duration

Excerpt:

Nautsch, Saeidi, Rathgeb, Busch Quality-based Score Calibration / Odyssey’16, Bilbao, 24.06.2016 3/3