Improving the Speaker Anonymization Evaluation's Robustness to Target Speakers with Adversarial Learning
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866915737327632384 |
|---|---|
| author | Franzreb, Carlos Das, Arnab Polzehl, Tim Möller, Sebastian |
| author_facet | Franzreb, Carlos Das, Arnab Polzehl, Tim Möller, Sebastian |
| contents | The current privacy evaluation for speaker anonymization often overestimates privacy when a same-gender target selection algorithm (TSA) is used, although this TSA leaks the speaker's gender and should hence be more vulnerable. We hypothesize that this occurs because the evaluation does not account for the fact that anonymized speech contains information from both the source and target speakers. To address this, we propose to add a target classifier that measures the influence of target speaker information in the evaluation, which can also be removed with adversarial learning. Experiments demonstrate that this approach is effective for multiple anonymizers, particularly when using a same-gender TSA, leading to a more reliable assessment. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_09803 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Improving the Speaker Anonymization Evaluation's Robustness to Target Speakers with Adversarial Learning Franzreb, Carlos Das, Arnab Polzehl, Tim Möller, Sebastian Audio and Speech Processing Machine Learning The current privacy evaluation for speaker anonymization often overestimates privacy when a same-gender target selection algorithm (TSA) is used, although this TSA leaks the speaker's gender and should hence be more vulnerable. We hypothesize that this occurs because the evaluation does not account for the fact that anonymized speech contains information from both the source and target speakers. To address this, we propose to add a target classifier that measures the influence of target speaker information in the evaluation, which can also be removed with adversarial learning. Experiments demonstrate that this approach is effective for multiple anonymizers, particularly when using a same-gender TSA, leading to a more reliable assessment. |
| title | Improving the Speaker Anonymization Evaluation's Robustness to Target Speakers with Adversarial Learning |
| topic | Audio and Speech Processing Machine Learning |
| url | https://arxiv.org/abs/2508.09803 |