---
license: mit
pipeline_tag: text-to-audio
---

# SCAPES

**SCAPES** (Semantically Conditioned Autoregressive Prior for Environmental Sounds) is a lightweight, resource-efficient generative model for synthesizing high-fidelity environmental textures through high-level semantic control.

This model is described in the paper [SCAPES: Semantically Conditioned Autoregressive Prior for Environmental Sounds](https://huggingface.co/papers/2609.04634).

- **Project page**: [https://cordutie.github.io/projects/scapes.html](https://cordutie.github.io/projects/scapes.html)
- **Code**: [https://github.com/cordutie/SCAPES](https://github.com/cordutie/SCAPES)