StreamSV

Code for the SPECOM 2026 paper "StreamSV: End-to-End Streaming Speaker Verification for Short Utterances."

Key configs

The three most important configurations, one per training stage:

• configs/wav2vec_vc12_long.yaml — offline utterance-level XLS-R extractor (enrollment).
• configs/w2v_incremental_2s_long.yaml — causal streaming prefix extractor.
• configs/w2v_incremental_scoring_i1.yaml — recurrent scoring head.

Checkpoint

Trained model checkpoint is available upon request.