Unsupervised adaptation of PLDA models for broadcast diarization

Viñals, Ignacio (Universidad de Zaragoza) ; Ortega, Alfonso (Universidad de Zaragoza) ; Villalba, Jesús ; Miguel, Antonio (Universidad de Zaragoza) ; Lleida, Eduardo (Universidad de Zaragoza)

Resumen: We present a novel model adaptation approach to deal with data variability for speaker diarization in a broadcast environment. Expensive human annotated data can be used to mitigate the domain mismatch by means of supervised model adaptation approaches. By contrast, we propose an unsupervised adaptation method which does not need for in-domain labeled data but only the recording that we are diarizing. We rely on an inner adaptation block which combines Agglomerative Hierarchical Clustering (AHC) and Mean-Shift (MS) clustering techniques with a Fully Bayesian Probabilistic Linear Discriminant Analysis (PLDA) to produce pseudo-speaker labels suitable for model adaptation. We propose multiple adaptation approaches based on this basic block, including unsupervised and semi-supervised. Our proposed solutions, analyzed with the Multi-Genre Broadcast 2015 (MGB) dataset, reported significant improvements (16% relative improvement) with respect to the baseline, also outperforming a supervised adaptation proposal with low resources (9% relative improvement). Furthermore, our proposed unsupervised adaptation is totally compatible with a supervised one. The joint use of both adaptation techniques (supervised and unsupervised) shows a 13% relative improvement with respect to only considering the supervised adaptation.

Idioma: Inglés

DOI: 10.1186/s13636-019-0167-7

Año: 2019

Publicado en: EURASIP JOURNAL ON AUDIO SPEECH AND MUSIC PROCESSING 2019, 24 (2019), [13 pp.]
ISSN: 1687-4714

Factor impacto JCR: 1.289 (2019)

Categ. JCR: ENGINEERING, ELECTRICAL & ELECTRONIC rank: 201 / 266 = 0.756 (2019) - Q4 - T3
Categ. JCR: ACOUSTICS rank: 21 / 32 = 0.656 (2019) - Q3 - T2

Factor impacto SCIMAGO: 0.289 - Electrical and Electronic Engineering (Q3) - Acoustics and Ultrasonics (Q3)

Financiación: info:eu-repo/grantAgreement/ES/DGA-FEDER/T36-17R
Financiación: info:eu-repo/grantAgreement/ES/MINECO/TIN2017-85854-C4-1-R

Tipo y forma: Artículo (Versión definitiva)

Área (Departamento): Área Teoría Señal y Comunicac. (Dpto. Ingeniería Electrón.Com.)