# EnCodec — AEmotionStudio mirror

Mirrors of Meta's [EnCodec](https://github.com/facebookresearch/encodec) checkpoints, used by Tessera v2 Latent mode (MAESTRO).

Upstream: https://github.com/facebookresearch/encodec  
License: MIT.

## Files

- `encodec_24khz.safetensors` — pre-RVQ encoder + decoder (safetensors via `save_model`, shared-tensor safe).
- `encodec_48khz.safetensors` — pre-RVQ encoder + decoder (safetensors via `save_model`, shared-tensor safe).

## Loading

```python
from encodec import EncodecModel
from safetensors.torch import load_model
model = EncodecModel.encodec_model_24khz(pretrained=False)
load_model(model, 'encodec_24khz.safetensors', strict=False)
model.eval()
```