Project Crying Angel · Sleeping AI

Listen. Rate. Advance music research.

A living human benchmark for music generated by artificial intelligence.

MusicalityBench

A benchmark that stays alive.

Models change. Music changes. Listeners change.

Created by Sleeping AI with leading music researchers, MusicalityBench continuously refreshes what people hear and evaluate. It is our first Active Benchmark and part of Project Crying Angel.

01New models
02New music
03New judgments
LICENCE

Before you contribute

Annotations, audio, listening coverage, coarse device and output categories, player volume, metadata, and other material submitted to this platform go directly to Sleeping AI’s research data systems. Sleeping AI owns the submitted platform data and may use it to train, fine-tune, augment, evaluate, and redistribute datasets or derived material to research and academic partners.

By submitting, you confirm that you have every permission and right required for the material. Sleeping AI assumes contributors are authorized to upload it and is not responsible for claims, losses, or disputes caused by unauthorized or unlawful submissions.

Identity required

Sign in and accept the data-use notice before a sample is issued.

Current evaluation

Blind listening session

Sign in to begin
1Listen
2Rate
3Submit

Blind sample

Authentication required

Model identity remains concealed
0% listenedDetecting playback context…

Your evaluation

Seven-point Likert assessment

Not at allVery much
Mood · GEMS framework · Required

Select every emotion the sample evokes.

Primary genre · Required

Select the closest fit; choose Other when needed.

Sign in to receive a sample.

Global participation

Where contributors come from.

0countries represented
Contributors by selected country

Secure contributor access

Continue with a passkey

Each contributor receives a unique research ID. No password, email, GPS location, or personal profile is required.

Quality control

Report this sample

Flag corrupt audio, unsafe content, or material that may lack the appropriate rights.