Datasets:
Formats:
parquet
Size:
10K - 100K
ArXiv:
Tags:
voice-cloning
text-to-speech
speaker-privacy
audio-protection
adversarial-audio
audio-deepfake
License:
File size: 740 Bytes
a932bd0 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 | # Protected LibriTTS references (RVCBench-Core v1)
Reference recordings from the LibriTTS portion of RVCBench after the protection methods evaluated in
RVC-AdvNoise and RVC-AntiProtect (paper Section 3.5 and Table 33). These are the 20 references used by
the Core v1 suite, taken unchanged from the runs reported in the paper.
| Folder | Method | Sample rate |
| --- | --- | --- |
| `gaussian` | Gaussian noise | 24 kHz |
| `spec` | SPEC | 24 kHz |
| `safespeech` | SafeSpeech | 24 kHz |
| `pop` | POP | 24 kHz |
| `enkidu` | Enkidu | 16 kHz |
| `demucs_spec` | SPEC references denoised with DEMUCS (dns64), for RVC-AntiProtect | 16 kHz |
Each file keeps the speaker folder and file name of its clean reference under `Libritts/audios/`.
|