• I went through different articles of speaker verifications (ECAPA-TDNN, TITANET).
  • They trained on VoxCeleb dataset.
  • In this dataset (voxceleb) some speakers have many more files than other speakers (imbalanced classes).
  • Those articles didn't mentioned how did they handle this imbalanced

Do we need to balanced the classes (number of files per speaker) ?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.