How We Trained Our AI on Ethically Sourced Drum Sounds (And Why It Matters)
The dirty details behind the data, what “ethical” looks like, and why it’s actually super important.
7 min read · Updated Jan 2026 · Written by real, human producers
Every time you read about an AI music tool, you'll see the phrase "royalty free." It's become a default claim — a checkbox ticked without explanation. But royalty free means nothing if the model generating your sounds was trained on music it had no right to use.
We think producers deserve to know what's actually going on under the hood. So here's ours.
Why training data is the real question
Generative AI learns by analyzing vast amounts of existing audio. The model studies patterns — the shape of transients, the character of frequencies, the relationship between attack and decay — and learns to generate new audio that reflects those patterns.
The critical question is: whose audio was it learning from?
If the answer is "music scraped from the internet without permission," then every sound the model generates carries the fingerprint of audio created by artists who never consented to being part of the training process. Whether that creates legal liability is still being debated in courts around the world. Whether it's ethical is not: it isn't.
How Just 4 Noise was trained
Our training data was assembled from audio that was either:
Created in-house — sounds built specifically for training purposes, with full ownership retained by Just 4 Noise
Licensed explicitly for AI training — audio sourced from creators and labels who provided written consent for use in generative model training
Public domain audio — historical recordings with fully expired copyright
We did not scrape. We did not assume that broad licenses meant AI training rights. We did not use Splice samples, YouTube audio, or any third-party library without explicit confirmation that AI training use was permitted.
This process was slower and more expensive than the alternative. We think it was the only acceptable approach.
This is also why we got confirmation and certification from AI for Music, an incredible organization backed by some of the biggest players in this space.
What this means for you
Legally: When you generate a drum sound with Just 4 Noise, you're generating something produced by a model trained on properly licensed audio. There's no copyright claim lurking in the training data. Use your sounds commercially, without reservation.
Creatively: Ethically sourced training data doesn't constrain what the AI can generate — it constrains how the model was built, not what it's capable of. The quality and range of sounds the model produces reflects the diversity and quality of the training data. We invested in good source material because we wanted a better model.
For the industry: Every time a producer chooses a tool with transparent, ethical training data over one that scraped without permission, it sends a signal. The music industry is watching how AI tools are built. The tools that survive long-term will be the ones that built trust from the start.
The transparency we're committing to
We'll continue to be open about how our models are trained as we expand the Just 4 Noise suite. When Just 4 Synths, Just 4 Trumpets, and Just 4 Fireworks launch, each model will have been trained to the same standard.
If you have specific questions about our training data or methodology, email us: info@just4noise.com. We'll answer.
Just 4 Kicks — ethically trained. Royalty free. Yours.
