SpeakerRPL v2 reduces equal error rate from 1.28% to 0.09% (93% relative reduction) on a Vox1-O-like test set by integrating reciprocal points learning with logit normalization, adaptive anchor learning, and model fusion.
GitHub - RVC-Boss/GPT-SoVITS: 1 min voice data can also be used to train a good TTS model! (few shot voice cloning)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.AS 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
SpeakerRPL v2: Robust Open-set Speaker Identification through Enhanced Few-shot Foundation Tuning and Model Fusion
SpeakerRPL v2 reduces equal error rate from 1.28% to 0.09% (93% relative reduction) on a Vox1-O-like test set by integrating reciprocal points learning with logit normalization, adaptive anchor learning, and model fusion.