---
license: apache-2.0
---

# MTIQ and MTSQ Datasets

## Overview

This dataset contains the preprocessed data used in the experiments reported in:
https://github.com/sb-lucas/Guided-Music-Retrieval

It consists of two related datasets:

- **MTIQ**: uses only original audio + triplet annotations  
- **MTSQ**: uses original audio + transformed audio subsets + triplet annotations

### Dataset Requirements

- To use **MTIQ**, download:
  - `original.tar.gz`
  - `triplets/MTIQ/`

- To use **MTSQ**, download:
  - `original.tar.gz`
  - `add_instrument.tar.gz`
  - `remove_instrument.tar.gz`
  - `mood_change.tar.gz`
  - `slowed.tar.gz`
  - `triplets/MTSQ/`

---

## Archive Contents

- `original.tar.gz`  
  Audio from the MTG-Jamendo dataset, trimmed to the central 10 seconds. Used as the base audio for both MTIQ and MTSQ.

- `add_instrument.tar.gz`  
  Audio subset corresponding to the *instrument addition* transformations in MTSQ.

- `remove_instrument.tar.gz`  
  Audio subset corresponding to the *instrument removal* transformations in MTSQ.

- `mood_change.tar.gz`  
  Audio subset corresponding to *mood change* transformationss in MTSQ.

- `slowed.tar.gz`  
  Audio subset corresponding to *velocity* transformations in MTSQ.

---

## Triplets

The `triplets/` directory contains precomputed train/validation/test splits for both MTIQ (`triplets/MTIQ/`) and MTSQ (`triplets/MTSQ/`), used for training and evaluation.