SpeakerCard-1M: Evidence-Grounded Speaker Card Corpus for Speaker Verification

Arxiv CS.SD (Sound/Audio ML)

SpeakerCard-1M dataset contains 1 million speaker cards with evidence-grounded natural language descriptions for in-the-wild speaker verification. Each card connects speaker embeddings to interpretable text, enabling natural-language queries for speaker identity.

speaker-verificationdatasets

Arxiv CS.SD (Sound/Audio ML)·Junyi PengOld\v{r}ich PlchotXiao SongDading ChongLichun FanHang SuThemos StafylakisJunjie LiKong Aik LeeShuai WangJan \v{C}ernock\'y

8 days ago