---
license: cc-by-4.0
task_categories:
  - audio-classification
tags:
  - piano
  - midi
  - performance-assessment
  - music
size_categories:
  - 1K<n<10K
---

# YCU-PPE-III: Piano Performance MIDI Dataset

MIDI transcriptions of the YCU-PPE-III piano performance dataset (Wang et al.), used for **unreferenced Performance MOS (PMOS) prediction** in [EVPMR](https://github.com/anusfoil/eval-piano-midi-repr).

## Overview

- **2,627 MIDI files** transcribed from WAV recordings via [transkun](https://github.com/RongTongZi/transkun)
- **13 songs** performed by student pianists
- **2,511 performances** with ratings from 3 expert judges (0-100 scale each)
- **Labels**: normalized mean score to [0, 1]
- **Splits**: 1,757 train / 377 val / 377 test (stratified by song)

## Metadata

`metadata/metadata.csv` columns: midi_path, name, song_id, piece_cn, piece_en, judge_1, judge_2, judge_3, mean_score, label, split

## Task

Regression: given a piano performance MIDI (without reference score), predict a quality rating.
Metrics: MSE, MAE, Pearson r, Spearman rho.

## Directory Structure

```
data/{song_id}/{name}.mid     # 2,627 MIDI files
metadata/splits.csv           # 2,511 entries: midi_path, label, split
metadata/metadata.csv         # Full metadata with judge scores + piece names
```