Final Year Project  ·  Human Evaluation Study

Novel Symbolic Music Generation Model

MusixLTM Listening Test Guide

Evaluating a new model for structured, high‑speed symbolic music generation — your ratings help assess its musical quality.

Before You Begin

Familiarise Yourself with MIDI

All clips in this study are rendered from MIDI — a format that may sound different from standard recordings. Listen to this sample track first to calibrate your ears before you start any test.

MIDI Sample

Quit Playing Games (With My Heart) — Backstreet Boys

0:00 —

Study Structure

Three Tests, One Session

You'll complete three online tests in sequence. Each takes approximately 15–20 minutes.

1

MOS Test

Quality
Rating

9 clips

Listen to each clip in full, then rate it across five musical quality criteria using a 1–5 scale from Bad to Excellent.

Structural Coherence Musical Flow +3 criteria
2

A/B Test

Side-by-Side
Comparison

9 pairs

Listen to two clips side by side and choose which you prefer across two specific dimensions.

Structural Coherence Overall Quality
3

Turing Test

Human
or AI?

10 clips

Listen to each clip and decide whether it was composed by a human or generated by an AI. Trust your instincts.

Human-composed AI-generated

Guidelines

Important Guidelines

  1. 🎧 Use Headphones
  2. ▶ Listen to Each Clip Fully Before Rating
  3. ✓ All Ratings Are Mandatory
  4. ⚠ Do Not Refresh the Page - It resets your progress !
  5. ← Use the "Previous" Button — Not Your Browser's Back Button

Consent Reminder

Before Each Test Begins

Each test opens with a consent page. By clicking "I Agree" you confirm:

You understand the purpose of the study and it has been explained to you.

Your responses will be recorded anonymously.

Your participation is entirely voluntary.