Low or No Background Audio
This one covers prerecorded audio with no picture, so podcasts and audiobooks rather than video. Where such a recording is mostly speech, one of three things has to be true. There is no background sound at all, or the listener can turn the background off, or the background runs at least 20 decibels below the voice. That last option forgives the occasional sound lasting a second or two. It does not forgive a continuous bed of them.
Why it matters
Twenty decibels is roughly four times quieter to the ear. Below that gap, hard of hearing listeners cannot pull the voice out of the music underneath it, and the recording turns into noise with words buried somewhere inside. The number is not arbitrary either. W3C took it from research on large-area assistive listening systems and on interference between hearing aids and digital phones.
Who this rule protects
Hard of hearing listeners are the group W3C names, and they are the group who cannot separate speech from background sound by concentrating harder.
How to check it yourself
- List your audio-only content. Podcasts, audiobooks, recorded statements. Leave video out, because this rule does not reach it.
- Skip anything that is mainly singing or rapping, along with audio logos and audio CAPTCHAs, all of which the criterion excludes.
- For what is left, check the mix for a music bed running under the whole piece rather than just the intro.
- Measure the separation rather than guessing at it, because 20 decibels is a number and a listening test is not.
- Look for a control that lets the listener mute the background on its own.
Failures we see most often
- A music bed sits just under the voice for energy, well short of the 20 decibel gap.
- Ambient sound stays at its recorded level for the whole episode.
- Short background sounds run one after another for minutes, which is not what the occasional-sound allowance covers.
Who this one is for
Read from this rule's own note above, so the grouping and the note cannot disagree.
- Deaf and hard of hearingpeople who cannot hear the audio
How to fix it
- Mix background sound at least 20 decibels below the speech, which is the one technique W3C documents here.
- Duck the music under the voice automatically instead of running it at a flat level throughout.
- Drop the bed entirely once the talking starts, which is the simplest of the three routes and needs no control at all.
- Where the background genuinely belongs to the piece, publish a speech-only version alongside it.
Step-by-step fix guides (1)
Passes vs. fails
Passes
The same episode drops the music once the host starts talking, and the show notes offer a speech-only cut.
Fails
A podcast runs its intro music at the same level as the host's voice for the first two minutes.
In audits and lawsuits
This is Level AAA and the only criterion in the standard that is a mixing decision rather than a code or design one, which makes it the cheapest of the lot to get right at production time and the easiest to forget. Its scope catches people out. Singing and rapping are carved out, so a song does not fail for having a band behind the vocal. So are audio logos and audio CAPTCHAs. Video is outside it altogether, which produces a strange result worth knowing. The same recording is in scope published as an MP3 and out of scope published as a video.