StreamVoiceAnon+: Emotion-Preserving Streaming Speaker Anonymization via Frame-Level Acoustic Distillation
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912947904708608 |
|---|---|
| author | Kuzmin, Nikita Lee, Kong Aik Chng, Eng Siong |
| author_facet | Kuzmin, Nikita Lee, Kong Aik Chng, Eng Siong |
| contents | We address the challenge of preserving emotional content in streaming speaker anonymization (SA). Neural audio codec language models trained for audio continuation tend to degrade source emotion: content tokens discard emotional information, and the model defaults to dominant acoustic patterns rather than preserving paralinguistic attributes. We propose supervised finetuning with neutral-emotion utterance pairs from the same speaker, combined with frame-level emotion distillation on acoustic token hidden states. All modifications are confined to finetuning, which takes less than 2 hours on 4 GPUs and adds zero inference latency overhead, while maintaining a competitive 180ms streaming latency. On the VoicePrivacy 2024 protocol, our approach achieves a 49.2% UAR (emotion preservation) with 5.77% WER (intelligibility), a +24% relative UAR improvement over the baseline (39.7%->49.2%) and +10% over the emotion-prompt variant (44.6% UAR), while maintaining strong privacy (EER 49.0%). Demo and code are available: https://anonymous3842031239.github.io/ |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_06079 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | StreamVoiceAnon+: Emotion-Preserving Streaming Speaker Anonymization via Frame-Level Acoustic Distillation Kuzmin, Nikita Lee, Kong Aik Chng, Eng Siong Audio and Speech Processing Artificial Intelligence Signal Processing We address the challenge of preserving emotional content in streaming speaker anonymization (SA). Neural audio codec language models trained for audio continuation tend to degrade source emotion: content tokens discard emotional information, and the model defaults to dominant acoustic patterns rather than preserving paralinguistic attributes. We propose supervised finetuning with neutral-emotion utterance pairs from the same speaker, combined with frame-level emotion distillation on acoustic token hidden states. All modifications are confined to finetuning, which takes less than 2 hours on 4 GPUs and adds zero inference latency overhead, while maintaining a competitive 180ms streaming latency. On the VoicePrivacy 2024 protocol, our approach achieves a 49.2% UAR (emotion preservation) with 5.77% WER (intelligibility), a +24% relative UAR improvement over the baseline (39.7%->49.2%) and +10% over the emotion-prompt variant (44.6% UAR), while maintaining strong privacy (EER 49.0%). Demo and code are available: https://anonymous3842031239.github.io/ |
| title | StreamVoiceAnon+: Emotion-Preserving Streaming Speaker Anonymization via Frame-Level Acoustic Distillation |
| topic | Audio and Speech Processing Artificial Intelligence Signal Processing |
| url | https://arxiv.org/abs/2603.06079 |