README.md

metadata

title: DeepAFx-ST
emoji: 🎛️
colorFrom: gray
colorTo: indigo
sdk: gradio
sdk_version: 6.24.0
python_version: '3.12'
app_file: app.py
pinned: false
license: other
license_name: adobe-research-license
short_description: Transfer a reference recording's production style
tags:
  - category:production
  - subcategory:mixing-style-transfer
  - input:audio
  - output:audio
  - output:file/json
  - style-transfer

HARP endpoint for DeepAFx-ST, an audio production style transfer system by Adobe Research. Given an input recording and a reference recording, the model applies differentiable parametric EQ and dynamic range compression to make the input match the reference's production style.

This deployment uses the autodiff variant trained on LibriTTS, loaded from the community checkpoint on Hugging Face. The code and model are released under the Adobe Research License for non-commercial research use only.

Source attribution is recorded in SOURCES.md.