Speech Content Privacy Enforcement on CitySpeechMix (evaluation)
9.4WER (%)Original audio
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Original audio2025.07 | 9.4 | 0 | 0 | |
| Cohen-Adria2025.07 | 83.6 | 25.1 | 21.8 | |
| Burkhardt2025.07 | 86.7 | 2.7 | 3.1 | |
| Segment-wise Waveform Reversal (Ours)mode=standard2025.07 | 97.9 | 2.7 | 1.4 | |
| Segment-wise Waveform Reversal (Ours)Speech Separation=SepFormer2025.07 | 97.9 | 1 | 2.7 | |
| Segment-wise Waveform Reversal (Ours)Speaker Identification=w/o SI2025.07 | 98 | 1.3 | 2.4 | |
| Segment-wise Waveform Reversal (Ours)Voice Activity Detection=w/o VAD2025.07 | 98.1 | 4.1 | 2.7 | |
| White noise2025.07 | 98.5 | 41 | 140.9 |