Background music for demo videos: licensing and the loudness math

"Royalty-free" is a billing term, not a permission slip#

The phrase that trips up every first demo is "royalty-free." It sounds like "free," and it is neither. Royalty-free means you pay once, or subscribe, and then owe no per-play royalty each time the video is watched. It says nothing about whether you owe a fee, a credit, or a license at all. A track can be royalty-free and still cost forty dollars, still demand attribution, still be forbidden in a product you sell. So the first job of scoring a music bed is not taste, it is reading the grant: who may use this, for what, where, and for how long.

For a product demo the answer to "for what" is almost always commercially. A demo sells something, which puts it outside the reach of a large, tempting slice of free music and inside a small set of licenses that actually permit it. Get that wrong and the cost is not a bad review. It is a Content ID claim that diverts your ad revenue to a stranger, or a takedown on the one video your launch depended on. Below is the map of what clears a commercial demo, and then the single number that decides whether the music helps or ruins it.

Six ways to license a bed, and what each really grants#

There are only a handful of legal routes to a music bed, and they differ less in sound than in the clause that bites you later. Where music sits in the end-to-end demo pipeline is a production question; this is the rights question underneath it.

Two rows carry most of the surprises. The Creative Commons NonCommercial and NoDerivatives variants read as free and are not usable for this: CC BY-NC 4.0 states plainly that "you may not use the material for commercial purposes," defined as anything "primarily intended for commercial advantage or monetary compensation"⁴. A paid product's marketing video is exactly that. The plain CC BY license is the friendly one: it permits use "for any purpose, even commercially," provided you give "appropriate credit, provide a link to the license, and indicate if changes were made"³. The YouTube Audio Library ships both flavors side by side; its filter tags each track "Attribution required" or not, and only the standard-license tracks are the ones YouTube vouches for as copyright-safe against Content ID⁵.

The subscription trap: the license is a lease, not a purchase#

The most expensive misreading is the subscription one, because it feels like buying and behaves like renting. Download a track from Epidemic Sound while subscribed, publish the demo, and that video stays cleared, even if you cancel next week. Publish a new video with the same downloaded file after your subscription lapses, and it "will result in copyright claims and monetization by Epidemic Sound"⁶. The audio file on your disk is not the thing you licensed; the right to publish with it is. Storyblocks draws the same line by tier: downloaded assets carry perpetual use only on its Business license, not on the Individual or Small Business plans⁷. The working rule is to keep a record of which track cleared which video under which active subscription, because "I downloaded it, so I own it" is the belief that ends in a claim.

If your demo carries live sound instead of an added bed, an app's own chime or a call tool ringing, that is a capture problem rather than a licensing one, and it lives in recording system audio and your voice together. This page is about the bed you lay under a finished narration.

The one number that separates amateur from pro: 20 dB#

Licensing decides whether the music is legal. One number decides whether it is any good: how far below the voice the bed sits. Amateur demos miss it in the same direction every time, mixing the music at or near the narration so the two fight and the words lose. The fix is not a feeling, it is a measured gap, and accessibility law has already fixed the figure.

WCAG's Success Criterion 1.4.7 requires background sound under foreground speech to be "at least 20 decibels lower than the foreground speech content," which it notes is "approximately four times quieter," with an exception only for occasional sounds lasting a second or two¹. It is a Level AAA criterion written for hard-of-hearing listeners, and the 20 dB value traces to hearing-aid and assistive-listening research, but it doubles as the mixing target for everyone: 20 dB is the separation at which a listener stops straining and the music turns into atmosphere instead of interference.

Loudness itself is measured in LUFS, Loudness Units relative to Full Scale, the scale standardized for broadcast and adopted worldwide. One LU equals one dB, and 0 LUFS is the top of the digital range, so real audio lives well below it in negative numbers. Broadcasters settled the absolute level long ago: EBU R128 pins a whole programme to an average of -23 LUFS, within half a decibel². Web and demo audio is usually mastered louder than that, commonly around -16 LUFS, because there is no shared transmission chain to protect. Anchor the voice near that master and the WCAG rule tells you exactly where the bed goes.

That table is the whole craft in three rows. The bed rides roughly 20 dB under the voice while a word is landing, and rises in the silence between phrases, because there is nothing to bury there.

Static ducking sets one level; dynamic ducking follows the voice#

There are two ways to enforce that gap, and the difference is audible. The static method picks one music level for the entire video and leaves it, say a flat 22 dB down. It never buries a word, but it also never breathes: parked that low, the bed is nearly inaudible in the pauses where it could add energy. The dynamic method, the one broadcast has used for decades, keys a compressor on the narration so the music dips the moment the voice enters and releases back up when it stops. The knobs are an attack fast enough to catch the first syllable, roughly 150 ms, and a release slow enough not to pump between words, roughly 500 to 700 ms.

Dynamic ducking is why a good demo's bed feels alive and a bad one feels like a light switch. Our own engine, aidemo (disclosed as ours), does this by default: the narration track sidechains a compressor on the music, so the bed swells over the intro card and ducks under every narrated line without anyone drawing a volume envelope, and a final normalization pass then masters the whole mix to -16 LUFS. The honest limits are the usual ones. It captures a browser only, its storyboard is written in code by an agent instead of assembled in a GUI, and it has no waveform editor for nudging a single duck by hand: you change a number in the spec and re-render. For a bed under a scripted voiceover that is enough, because ducking is arithmetic, not taste, and arithmetic belongs in a config file.

When a bed earns its keep, and when silence wins#

Music is not free even when the license is. It adds a track to balance, a claim to track, and one more way for a render to fail. So the last call is whether the demo needs a bed at all. A launch film, a social cut, a Product Hunt hero: yes, energy carries them and silence reads as cheap. A thirty-second feature clip in a docs page, a muted autoplay loop, an onboarding walkthrough someone plays at their desk: often no, because the value is the narration and the UI, and every decibel of music is a decibel competing with the point. Ship most demos with the voice clean and the option to add a bed later, and save the music for the cuts where mood is the job. When you do add it, the failure to dodge is the eighth-costliest error in the ranked catalog of what loses viewers: a bed mixed over the voice. Twenty decibels down and keyed to the narration, the music does its one job, which is to be felt and never heard over the synthetic voice it rides under.

Sources#

1. W3C — Understanding SC 1.4.7: Low or No Background Audio, WCAG 2.2 (20 dB rule)
2. EBU — R 128: Loudness normalisation and permitted maximum level of audio signals (-23 LUFS)
3. Creative Commons — CC BY 4.0 deed (commercial use permitted with attribution)
4. Creative Commons — CC BY-NC 4.0 deed (no commercial use)
5. YouTube Help — Use music and sound effects from the Audio Library
6. Epidemic Sound — How the music license works (rights after cancellation)
7. Storyblocks — License comparison (perpetual use by tier)

FAQ#

Do I need a license for background music in a product demo?#

Yes. A product demo is commercial use, so you need a license that clears commercial use, not just any free track. That rules out Creative Commons NonCommercial and NoDerivatives music outright, and it means "royalty-free" is not enough on its own, since royalty-free only waives per-play royalties, not the license itself. Safe routes are CC0 or public-domain tracks, CC BY with proper credit, the YouTube Audio Library, or a paid subscription or per-track library whose terms name commercial use.

How loud should background music be under a voiceover?#

At least 20 decibels below the voice while someone is speaking, which WCAG 1.4.7 describes as roughly four times quieter. In loudness terms, if the narration is mastered near -16 LUFS, the ducked bed sits near -36 LUFS under the words and is free to swell back up in the gaps between phrases. Enforce it with dynamic sidechain ducking rather than a single fixed level, so the music dips only when the voice enters and breathes in the silences.

Can I use Epidemic Sound or Artlist music after I cancel?#

Only in what you already published. Subscription libraries license the act of publishing, not the downloaded file: Epidemic Sound keeps videos you published during an active subscription cleared forever, but any new video you release after you cancel gets copyright claims. Storyblocks grants perpetual use of downloaded assets only on its Business tier, not the cheaper plans. Keep a log of which track cleared which video under which live subscription so you can prove it later.