README.md

---
title: Audio Difficulty Estimator
emoji: 🎹
colorFrom: purple
colorTo: pink
sdk: gradio
sdk_version: "4.26.0"
python_version: 3.10.13
app_file: app.py
pinned: false
tags:
  - music
  - audio
  - piano
  - difficulty-estimation
short_description: Estimate piano difficulty from audio
hardware: "a100-large"
---

# 🎼 Music Difficulty Estimator

This Gradio app estimates the **difficulty of piano pieces** based on uploaded audio (MP3/MP4) or YouTube links. It uses pretrained models to generate a MIDI transcription and predict difficulty from three musical perspectives:

- CQT-based representation
- Piano roll representation
- Multimodal embeddings

## 🛠 How it works

1. You upload an audio or video file, or paste a YouTube link.
2. The audio is transcribed to MIDI using a piano transcription model.
3. Three different difficulty models analyze the audio and generate predictions.
4. You can listen to the extracted MP3 and the generated MIDI.

## 📦 Model loading

All models are stored separately in the [pramoneda/audio](https://huggingface.co/pramoneda/audio) model repository and are downloaded dynamically via `huggingface_hub`.

## 📁 Input formats

- MP3 audio
- MP4 video (audio extracted automatically)
- YouTube links

## ✨ Built with

- `gradio` for the interface
- `pydub` and `yt_dlp` for audio processing
- `huggingface_hub` to load model checkpoints
- `ffmpeg-python` for format conversion

## 🔗 Related

- [Model repo: pramoneda/audio](https://huggingface.co/pramoneda/audio)
- [More projects by pramoneda](https://huggingface.co/pramoneda)

---