Large-scale per-speaker evaluation shows re-identification risk in speech anonymization arises from attacker-anonymizer-data interactions, challenging notions of intrinsic speaker privacy levels.
2nd V oicePrivacy Challenge Evaluation Plan
7 Pith papers cite this work. Polarity classification is still indexing.
7
Pith papers citing it
representative citing papers
SRD provides a threshold-independent, representation-level privacy assessment for voice anonymization that reveals system weaknesses not detected by equal error rate evaluation.
A dual-branch multimodal model combining ECAPA-TDNN on anonymized audio and BERT on transcripts outperforms prior attackers on five of seven VPAC benchmarks and reaches SOTA with augmentation.
Listeners detect automatic anonymization in pathological speech at 91-93% accuracy with a 30-point perceived quality drop, yet clinical severity ratings stay nearly unchanged for dysarthria, dysglossia, and dysphonia.
TF-MoE uses dynamic per-frame and per-mel-band expert selection in time and frequency dimensions to improve speech separation performance at comparable compute cost to prior models.
A self-supervised prosody encoder with speaker disentanglement strategies outperforms raw prosody and HuBERT baselines on pitch reconstruction and prosodic event detection while achieving strong speaker separation.
A two-stage framework replaces personally identifiable information via generative editing and anonymizes voices with a flow-matching model to achieve stronger privacy than VoicePrivacy baselines while keeping utility high for retrained ASR, TTS, and SER models.
citing papers explorer
-
A Large-Scale Per-Speaker Analysis of Re-identification Risk in Speech Anonymization
Large-scale per-speaker evaluation shows re-identification risk in speech anonymization arises from attacker-anonymizer-data interactions, challenging notions of intrinsic speaker privacy levels.
-
Evaluating voice anonymisation using similarity rank disclosure
SRD provides a threshold-independent, representation-level privacy assessment for voice anonymization that reveals system weaknesses not detected by equal error rate evaluation.
-
VoxATtack: A Multimodal Attack on Voice Anonymization Systems
A dual-branch multimodal model combining ECAPA-TDNN on anonymized audio and BERT on transcripts outperforms prior attackers on five of seven VPAC benchmarks and reaches SOTA with augmentation.
-
Perceptual implications of automatic anonymization in pathological speech
Listeners detect automatic anonymization in pathological speech at 91-93% accuracy with a 30-point perceived quality drop, yet clinical severity ratings stay nearly unchanged for dysarthria, dysglossia, and dysphonia.
-
TF-MoE: Time-Frequency Mixture-of-Experts for Efficient Speech Separation
TF-MoE uses dynamic per-frame and per-mel-band expert selection in time and frequency dimensions to improve speech separation performance at comparable compute cost to prior models.
-
Privacy-preserving Prosody Representation Learning
A self-supervised prosody encoder with speaker disentanglement strategies outperforms raw prosody and HuBERT baselines on pitch reconstruction and prosodic event detection while achieving strong speaker separation.
-
Anonymization, Not Elimination: Utility-Preserved Speech Anonymization
A two-stage framework replaces personally identifiable information via generative editing and anonymizes voices with a flow-matching model to achieve stronger privacy than VoicePrivacy baselines while keeping utility high for retrained ASR, TTS, and SER models.