Edit Content, Preserve Acoustics: Imperceptible Text-Based Speech Editing via Self-Consistency Rewards
Paper • 2602.00560 • Published
wavepainter
Released checkpoints for wavepainter, a masked-span speech editor.
On the English full split of Ming-Freeform-Audio-Edit, against Ren et al.
(arXiv:2602.00560, the row with GRPO):
| edit type | metric | Ren et al. | wavepainter |
|---|---|---|---|
| substitution | WER | 4.41 | 3.063 |
| insertion | WER | 4.97 | 4.128 |
| deletion | WER | 6.88 | 9.136 |
| substitution | SIM | 0.78 | 0.943 |
| insertion | SIM | 0.82 | 0.961 |
| deletion | SIM | 0.78 | 0.918 |
| substitution | DNSMOS | 3.11 | 2.931 |
| insertion | DNSMOS | 3.18 | 2.905 |
| deletion | DNSMOS | 3.09 | 2.938 |
Five of the nine numbers they report improve on theirs: substitution and insertion WER, and speaker similarity on all three edit types. We do not beat them on deletion WER or on DNSMOS anywhere -- see the repository's Limitations.
Files
phase1-base/— the phase-1 editor. Start here to re-run phase 2 (~1 GPU-hour).phase2-hubert-a085/— the released model. This produces the numbers above.
Use
git clone https://github.com/pujariaditya/wavepainter && cd wavepainter
pip install -e . && ./setup.sh
python scripts/download_weights.py
./scripts/verify_benchmark.sh
Checkpoints are sha256-pinned in scripts/download_weights.py. Licence and
attribution, including for the components these weights derive from, are in the
repository's ATTRIBUTION.md.