Quantifying Source Speaker Leakage in One-to-One Voice Conversion
arXiv:2504.15822 · doi:10.1109/BIOSIG61931.2024.10786731
Abstract
Using a multi-accented corpus of parallel utterances for use with commercial speech devices, we present a case study to show that it is possible to quantify a degree of confidence about a source speaker's identity in the case of one-to-one voice conversion. Following voice conversion using a HiFi-GAN vocoder, we compare information leakage for a range speaker characteristics; assuming a "worst-case" white-box scenario, we quantify our confidence to perform inference and narrow the pool of likely source speakers, reinforcing the regulatory obligation and moral duty that providers of synthetic voices have to ensure the privacy of their speakers' data.
arXiv commentAccepted at IEEE 23rd International Conference of the Biometrics Special Interest Group (BIOSIG 2024)
References in corpus (5)
- VoxCeleb2: Deep Speaker Recognition
- ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification
- HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis
- A Comparison of Discrete and Soft Speech Units for Improved Voice Conversion
- Evaluating Voice Conversion-based Privacy Protection against Informed Attackers