Training Speaker Embedding Extractors Using Multi-Speaker Audio with Unknown Speaker Boundaries

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216305%3A26230%2F22%3APU146060" target="_blank" >RIV/00216305:26230/22:PU146060 - isvavai.cz</a>

Result on the web

<a href="https://www.isca-speech.org/archive/pdfs/interspeech_2022/stafylakis22_interspeech.pdf" target="_blank" >https://www.isca-speech.org/archive/pdfs/interspeech_2022/stafylakis22_interspeech.pdf</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.21437/Interspeech.2022-10165" target="_blank" >10.21437/Interspeech.2022-10165</a>

Result language

angličtina

Original language name

Training Speaker Embedding Extractors Using Multi-Speaker Audio with Unknown Speaker Boundaries

Original language description

In this paper, we demonstrate a method for training speaker em- bedding extractors using weak annotation. More specifically, we are using the full VoxCeleb recordings and the name of the celebrities appearing on each video without knowledge of the time intervals the celebrities appear in the video. We show that by combining a baseline speaker diarization algorithm that re- quires no training or parameter tuning, a modified loss with aggregation over segments, and a two-stage training approach, we are able to train a competitive ResNet-based embedding extractor. Finally, we experiment with two different aggregation functions and analyze their behaviour in terms of their gradients.

Czech name

—

Czech description

—

Type

D - Article in proceedings

CEP classification

—

OECD FORD branch

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Project

Result was created during the realization of more than one project. More information in the Projects tab.

Continuities

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Publication year

2022

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Article name in the collection

Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH

ISBN

—

ISSN

1990-9772

e-ISSN

—

Number of pages

5

Pages from-to

605-609

Publisher name

International Speech Communication Association

Place of publication

Incheon

Event location

Incheon Korea

Event date

Sep 18, 2022

Type of event by nationality

WRD - Celosvětová akce

UT code for WoS article

—