Joon‐Hyuk Chang

Improving Transformer-based End-to-End Speaker Diarization by Assigning Auxiliary Losses to Attention Heads

2023 · arXiv (Cornell University)

Transformer-based end-to-end neural speaker diarization (EEND) models utilize the multi-head self-attention (SA) mechanism to enable accurate speaker label prediction in overlapped speech regions. In this study, to enhance the training effectiveness of SA-EEND models, we …

Trainable Adaptive Score Normalization for Automatic Speaker Verification

2025

Adaptive S-norm (AS-norm) calibrates automatic speaker verification (ASV) scores by normalizing them utilize the scores of impostors which are similar to the input speaker. However, AS-norm does not involve any learning process, limiting its ability …