---
language:
- en
license: cc-by-4.0
pretty_name: Free Music Archive (FMA) Dataset
size_categories:
- 1K<n<10K
source_datasets:
- original
task_categories:
- audio-classification
task_ids:
- multi-class-classification
---

# Free Music Archive (FMA) Dataset

## Overview

This repository contains the Free Music Archive (FMA) dataset, curated and made available on Hugging Face by [dragunflie-420](https://huggingface.co/dragunflie-420). The FMA dataset is a large-scale, open-source dataset of music tracks, designed for music information retrieval and machine learning tasks.

## Dataset Description

The Free Music Archive (FMA) is an open and easily accessible dataset consisting of full-length audio tracks with associated metadata. This particular version focuses on the "small" subset of the FMA, which includes:

- 8,000 tracks of 30 seconds each
- 8 balanced genres (Electronic, Experimental, Folk, Hip-Hop, Instrumental, International, Pop, Rock)
- Audio files in 128k MP3 format
- Comprehensive metadata for each track

## Contents

This dataset provides:

1. Audio files: 30-second MP3 clips of music tracks
2. Metadata: Information about each track, including:
   - Track ID
   - Title
   - Artist
   - Genre
   - Additional features (e.g., acoustic features, music analysis data)

## Data Files

To use this dataset, you need to manually download and place the following files in the repository:

1. `fma_small.zip`: Contains the audio files
2. `fma_metadata.zip`: Contains the metadata for the tracks

After downloading, extract these files and ensure the following directory structure:

```
fma_dataset/
├── fma_small/
│   ├── 000/
│   ├── 001/
│   └── ...
└── fma_metadata/
    ├── tracks.csv
    ├── genres.csv
    └── features.csv
```

## Usage

To use this dataset in your Hugging Face projects:

```python
from datasets import load_dataset

dataset = load_dataset("dragunflie-420/fma")

# Access the first example
first_example = dataset['train'][0]
print(first_example['title'], first_example['artist'], first_example['genre'])

# Play the audio (if in a notebook environment)
from IPython.display import Audio
Audio(first_example['audio']['array'], rate=first_example['audio']['sampling_rate'])
```

[... rest of the README content remains the same ...]





---
language:
- en
license: cc-by-4.0
pretty_name: Free Music Archive (FMA) Dataset
size_categories:
- 1K<n<10K
source_datasets:
- original
task_categories:
- audio-classification
task_ids:
- multi-class-classification
---

# Free Music Archive (FMA) Dataset

## Overview

This repository contains the Free Music Archive (FMA) dataset, curated and made available on Hugging Face by [dragunflie-420](https://huggingface.co/dragunflie-420). The FMA dataset is a large-scale, open-source dataset of music tracks, designed for music information retrieval and machine learning tasks.

[... rest of the README content remains the same ...] Free Music Archive (FMA) Dataset

## Dataset Description

The Free Music Archive (FMA) is an open and easily accessible dataset consisting of full-length audio tracks with associated metadata. This particular version focuses on the "small" subset of the FMA, which includes:

- 8,000 tracks of 30 seconds each
- 8 balanced genres (Electronic, Experimental, Folk, Hip-Hop, Instrumental, International, Pop, Rock)
- Audio files in 128k MP3 format
- Comprehensive metadata for each track

## Contents

This dataset provides:

1. Audio files: 30-second MP3 clips of music tracks
2. Metadata: Information about each track, including:
   - Track ID
   - Title
   - Artist
   - Genre
   - Additional features (e.g., acoustic features, music analysis data)

## Usage

To use this dataset in your Hugging Face projects:

```python
from datasets import load_dataset

dataset = load_dataset("dragunflie-420/fma")

# Access the first example
first_example = dataset['train'][0]
print(first_example['title'], first_example['artist'], first_example['genre'])

# Play the audio (if in a notebook environment)
from IPython.display import Audio
Audio(first_example['audio']['array'], rate=first_example['audio']['sampling_rate'])
```

## Dataset Structure

Each example in the dataset contains:

- `track_id`: Unique identifier for the track
- `title`: Title of the track
- `artist`: Name of the artist
- `genre`: Top-level genre classification
- `audio`: Audio file in the format compatible with Hugging Face's Audio feature

## Applications

This dataset is suitable for various music information retrieval and machine learning tasks, including:

- Music genre classification
- Artist identification
- Music recommendation systems
- Audio feature extraction and analysis
- Music generation and style transfer

## Citation

If you use this dataset in your research, please cite the original FMA paper:

```
@inproceedings{defferrard2016fma,
  title={FMA: A Dataset for Music Analysis},
  author={Defferrard, Micha{\"e}l and Ben