Voice Anonymization on LibriSpeech (dev)
1.8WERGround Truth
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ground TruthProsody (wpro)=–, Speaker=–2026.04 | 1.8 | — | — | — | |
| T9*Prosody (wpro)=–, Speaker=–2026.04 | 2.33 | — | 33.43 | — | |
| B1*Prosody (wpro)=–, Speaker=pseudo2026.04 | 3.07 | — | 9.2 | — | |
| T8*Prosody (wpro)=–, Speaker=–2026.04 | 3.45 | — | 40.93 | 49.22 | |
| T10*Prosody (wpro)=–, Speaker=–2026.04 | 3.51 | — | 42.45 | — | |
| B3Prosody (wpro)=–, Speaker=pseudo2026.04 | 4.16 | 44.08 | 25.74 | 37.49 | |
| B5Prosody (wpro)=–, Speaker=pseudo2026.04 | 4.81 | 48.68 | 34.37 | 66.64 | |
| DiffAnonProsody (wpro)=1.0, Speaker=ψ2026.04 | 4.91 | 35.23 | 14.97 | 76.67 | |
| DiffAnonProsody (wpro)=0.8, Speaker=ψ2026.04 | 5.11 | 37.74 | 15.73 | 74.7 | |
| DiffAnon + pitch-shiftProsody (wpro)=1.0, Speaker=ψ2026.04 | 5.21 | 39.02 | 16.47 | 74.36 | |
| DiffAnonProsody (wpro)=0.5, Speaker=ψ2026.04 | 5.44 | 41.74 | 19.57 | 69.56 | |
| DiffAnonProsody (wpro)=0.2, Speaker=ψ2026.04 | 5.73 | 43.91 | 22.4 | 65.58 | |
| DiffAnonProsody (wpro)=null, Speaker=ψ2026.04 | 5.79 | 45.03 | 23.82 | 64.32 | |
| DiffAnonProsody (wpro)=null, Speaker=null2026.04 | 5.92 | 44.86 | 22.49 | 64.09 | |
| B4Prosody (wpro)=–, Speaker=pseudo2026.04 | 6.12 | 50.14 | 32.89 | 70.93 | |
| DiffAnonProsody (wpro)=null, Speaker=ψ, spk CFG (wspk = 3)2026.04 | 6.63 | 48.44 | 27.37 | 57.05 | |
| B6*Prosody (wpro)=–, Speaker=pseudo2026.04 | 9.69 | — | 23.05 | — | |
| B2Prosody (wpro)=–, Speaker=pseudo2026.04 | 10.48 | 31.52 | 8.45 | 58.7 |