# VoicePostprocessModel Checkpoints

Trained checkpoints for the DACVAE/PCA recipe in VoicePostprocessModel.

Files:
- `dacvae_estimator_best.pt`: differentiable acoustic-state estimator over Echo-TTS Fish/DAC + PCA latents.
- `dacvae_editor_general_best.pt`: broad dry-recovery editor trained on room, EQ, distortion, compression, bandwidth, and noise variants.
- `dacvae_editor_target_reverb_best.pt`: target-only editor trained on the calibrated room/bathroom reverb suite.
- `dacvae_editor_mixed_corrective_best.pt`: mixed corrective editor fine-tuned on the calibrated suite plus the broad cache.

Latent space dependencies:
- Fish/DAC checkpoint: https://huggingface.co/jordand/fish-s1-dac-min
- Echo-TTS PCA state: https://huggingface.co/jordand/echo-tts-base/blob/main/pca_state.safetensors