CopyPaste-based parallel data, cosine similarity loss, and energy-based emotion masking reduce speaker verification EER by 19.29% relative on the Dusha emotional speech corpus.
Datasets Our models are first pre-trained on the V oxCeleb [22] and then fine- tuned on the Dusha [18] dataset to evaluate the performance of SV
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
baseline 1
citation-polarity summary
fields
cs.SD 1years
2025 1verdicts
CONDITIONAL 1roles
baseline 1polarities
baseline 1representative citing papers
citing papers explorer
-
Learning Emotion-Invariant Speaker Representations for Speaker Verification
CopyPaste-based parallel data, cosine similarity loss, and energy-based emotion masking reduce speaker verification EER by 19.29% relative on the Dusha emotional speech corpus.