README.md

---
license: cc-by-nc-4.0
task_categories:
- audio-classification
tags:
- texture-sounds
pretty_name: Micro Texture Sounds Datase
---

# 📦 Sound Texture Dataset Collection

This dataset is a collection of three distinct sound texture datasets, prepared for machine learning tasks such as classification, generation, or analysis of audio textures. Each subset comes from different sources and includes metadata where applicable.

---

## 📁 Dataset Structure

The repository contains the following folders:

### 1. `boreillysegmented16K_class/`

This folder contains the **BOReillySegmented16K** dataset, based on the original BOReilly collection:

- 📎 **Samples**: 6580 one-second audio clips  
- 🎧 **Sample Rate**: 16kHz  
- 📁 **File Size**: ~204MB (zipped)  
- 🔊 **Content**: Loosely curated noisy sound textures  
- 🧑‍🎨 **Source**: Sound artist Brian O'Reilly  
- 📄 **Original Use**: Featured in a paper submitted to evoMusArt  
- ⚠️ **Note**: The folder does **not** include the `test` subset from the original dataset  
- ⏱ **Duration**: 6580 seconds, divided into **10 classes**

> For more info on the original dataset, visit [BOReilly Dataset Page](https://www.evomusart.org/).

---

### 2. `freesound_class/`

A curated subset of sounds sourced from [Freesound.org](https://freesound.org/):

- ✅ All samples are licensed under **Creative Commons licenses** at least as permissive as **CC BY-NC 4.0**  
- 📂 Includes a `metadata.json` with links to each original sound and license info  
- 🛠️ Hand-curated to focus on sound **textures**
- 🎧 **Sample Rate**: 44100Hz  
- 🖤 **Mono**  
- ⏱ **Duration**: 2261 seconds, divided into **6 classes**

---

### 3. `syntex_class/`

A small, synthetic dataset generated using [SynTex](https://nime.org/proc/nime22_15/), a sound texture generator tool:

- 🔬 Generated using select sound texture models  
- 📚 Dataset originally published in the paper:  
  **"SynTex: Generating Audio Texture Datasets"**  
  DOI: [10.21428/92fbeb44.0fe70450](https://doi.org/10.21428/92fbeb44.0fe70450)  
- 🌐 Full paper: [NIME 2022 Proceedings](https://nime.org/proc/nime22_15/)
- 🎧 **Sample Rate**: 44100Hz  
- 🖤 **Mono**  
- ⏱ **Duration**: 7286 seconds, divided into **6 classes**

---

## 📑 License

- `boreillysegmented16K_class`: Refer to usage terms from the original authors / evoMusArt submission.
- `freesound_class`: Sounds are under Creative Commons licenses (see metadata).
- `syntex_class`: Follows the licensing and terms outlined in the original SynTex paper.

Please ensure appropriate attribution if using this dataset in your work.

---

## 📬 Citation

If you use this dataset collection in your work, consider citing the original datasets or papers associated with each subset.