Spaces:
Running on Zero
Running on Zero
File size: 1,406 Bytes
3e70344 f55264b 2c685c1 3e70344 5074382 f55264b 3e70344 0bd19ce 5c15765 74b77b4 5c15765 f55264b 3e70344 f55264b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 | ---
title: DExter
emoji: 🎹
colorFrom: purple
colorTo: indigo
sdk: gradio
sdk_version: 6.24.0
python_version: '3.10'
app_file: app.py
pinned: false
license: mit
short_description: Render expressive piano performance from a score
tags:
- category:synthesis
- subcategory:performance-rendering
- domain:music
- input:file/xml|musicxml
- output:midi
- piano
startup_duration_timeout: 1h
---
# DExter
HARP endpoint for expressive piano performance rendering. Upload a MusicXML piano score, or use the bundled Schubert example, and download a generated MIDI performance. The model uses the official score-only pretrained checkpoint. Perceptual feature controls and style transfer are not available in the official custom-score inference path.
The upstream source code is MIT licensed. The separately hosted checkpoint has no stated redistribution license, so the Space is marked with an unknown overall license; see [SOURCES.md](SOURCES.md).
## Local setup
```bash
uv venv --python 3.10 .venv
UV_CACHE_DIR=/path/to/writable/cache uv pip install --python .venv/bin/python -r requirements.txt
git lfs pull
.venv/bin/python app.py
```
Run direct inference with `from runtime import render; render("score.musicxml", "performance.mid")`. The official sampler uses 999 diffusion steps and can be slow on CPU. Source, checkpoint provenance, and redistribution note are in [SOURCES.md](SOURCES.md).
|