Speaker verification improves with multi-corpus training and data scaling
ReDimNet2+: Multi-Corpus Data Scaling for Robust Speaker Verification
Sound
Summary
Speaker verification systems check if a voice matches a specific person, but they can struggle with different recording devices or environments. The authors created ReDimNet2+, a method trained on many speech collections to make voice recognition more reliable under varied conditions. They combined techniques like sound data changes and fine-tuning to lower error rates significantly. Their approach also includes smart ways to pick the best voice matches from a large library.
What this means in practice
- •For voice biometrics teams: Improve speaker ID systems across many devices and environments by training on diverse speech datasets and using advanced fine-tuning and reranking.
- •For security system developers: Enhance robustness of speaker verification in complex conditions like phone compression and noisy rooms by incorporating novel data augmentation and model training techniques.
Authors
Kirill Borodin, Vasilii Kudryavtsev, Maxim Maslov, Grach Mkrtchian
Abstract
Automatic speaker verification must remain reliable across devices, rooms, and compression pipelines. We present ReDimNet2+, which scales training of the compact ReDimNet2 backbone across seven public corpora (63,934 speakers, about 8,675 hours). Analysis of a VoxBlink2 subset reveals a shift in predicted spectral coloration, motivating codec and waveform augmentation alongside this multi-corpus training, large-margin fine-tuning (LMFT), and graph-based retrieval reranking. With random 4-second evaluation windows for all models, ReDimNet2+ LMFT reduces pooled VoxCeleb1 EER from 2.42% to 0.82% and a 26-condition robustness stress-test EER from 7.21% to 1.99%. Under this shared local protocol, it reaches 0.35% EER on VoxCeleb1-O versus 0.787% for the best evaluated WeSpeaker checkpoint. On a VoxBlink2 retrieval subset, reranking improves the final model's Pr@k from 0.7413 to 0.7687.