DisSR: Disentangling Speech Representation for Degradation-Prior Guided Cross-Domain Speech Restoration
Anonymous
Pipeline
Evaluation Results on Band-limited distoration (LibriTTS -> VCTK)
| Distorted | VoiceFixer | SelfRemaster | DisSR | Ground-Truth |
|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Evaluation Results on Clipped distoration (LibriTTS -> VCTK)
| Distorted | VoiceFixer | SelfRemaster | DisSR | Ground-Truth |
|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Evaluation Results on Quantized&Resampled distoration (LibriTTS -> VCTK)
| Distorted | VoiceFixer | SelfRemaster | DisSR | Ground-Truth |
|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Evaluation Results on Overdrive distoration (LibriTTS -> VCTK)
| Distorted | VoiceFixer | SelfRemaster | DisSR | Ground-Truth |
|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Other comments
1. We share our code on github, which aims to experimental reproduction and verification.
2. We are internally confirming the copyright issues of the model and data, after which the pretrained models will be released.