Skip to main navigation Skip to search Skip to main content

ECAPA-TDNN Embeddings for Speaker Recognition

  • Jingxiang Guo
  • , Jinxuan Zhu
  • , Sixu Lin
  • , Feng Shi*
  • *Corresponding author for this work
  • Harbin Institute of Technology Shenzhen
  • Harbin Institute of Technology

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

This paper proposes novel robust speaker recognition pipelines to tackle the problem of far-field speaker recognition in reverberation and noise to the Robovox dataset. Specifically, we explored methods including I-vector, X-vector, and ECAPA-TDNN, with the ECAPA-TDNN method demonstrating exceptional performance. The paper details the comprehensive process of our preparation and experiment.

Original languageEnglish
Title of host publication2024 5th International Seminar on Artificial Intelligence, Networking and Information Technology, AINIT 2024
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages1488-1491
Number of pages4
ISBN (Electronic)9798350385557
DOIs
StatePublished - 2024
Externally publishedYes
Event5th International Seminar on Artificial Intelligence, Networking and Information Technology, AINIT 2024 - Hybrid, Nanjing, China
Duration: 29 May 202431 May 2024

Publication series

Name2024 5th International Seminar on Artificial Intelligence, Networking and Information Technology, AINIT 2024

Conference

Conference5th International Seminar on Artificial Intelligence, Networking and Information Technology, AINIT 2024
Country/TerritoryChina
CityHybrid, Nanjing
Period29/05/2431/05/24

Keywords

  • Aggregation in TDNN (ECAPA-TDNN)
  • Emphasized Channel Attention
  • Propagation
  • Speaker Recognition
  • Time Delay Neural Networks (TDNN)

Fingerprint

Dive into the research topics of 'ECAPA-TDNN Embeddings for Speaker Recognition'. Together they form a unique fingerprint.

Cite this