The number
Mix the gap, not the music
Asking how loud the music should be is the wrong question, because the answer depends entirely on how loud your voice is. What you are actually setting is the distance between the two.
For guided audio that gap wants to be roughly 12 to 20 dB. Voice sits forward and steady; music sits underneath it, present enough that the silences feel held but never competing for attention. Below about 12 dB of separation the music starts to mask consonants, and the listener does the tiring work of decoding rather than following.
For the finished file, aim somewhere around −18 LUFS integrated. That is the loudness meditation apps tend to normalise to, so mastering much hotter buys you nothing — the platform simply turns it back down, and you have spent your dynamic range for no gain.
Set your voice level first and leave it alone. Then bring the music up from silence until you can just feel it, and stop there. Mixing in that order stops the slow creep where each pass makes the bed a little louder.
By format
How far under depends on how much you're asking of the listener
The more the listener has to actively follow, the quieter the bed. Instruction-dense work needs headroom for the words; sessions that dissolve into silence can let the music carry more.
- Instructional or technique-led: 18–20 dB under voice
- General guided meditation: 15–18 dB under
- Body scan and somatic work: 15–20 dB, filling the dwell time
- Breathwork with a count: 15–18 dB, and never let it obscure the cue
- Sleep: 12–15 dB, rising as the voice drops away
- Yoga nidra: 18–20 dB — the rotation needs clear words
The test that catches it
Listen to the silences, not the speech
A music bed almost always sounds fine underneath talking. The problem shows up in the gaps, where the music is suddenly alone and reveals how loud it really is. If the bed appears to jump forward every time you stop speaking, it is too loud, even though nothing about the speech sounded wrong.
This matters more in guided audio than in any other spoken format, because the pauses are long and deliberate. A meditation is mostly silence by design, so the music is exposed for most of the running time.
Check the mix twice: once on earbuds and once through a phone speaker at low volume. The phone speaker is unforgiving and it is what a lot of people will actually use at bedtime. If the words survive there, they will survive anywhere.
Avoid automatic ducking that dips the music every time you speak. On a track with long pauses it pumps audibly, and the movement pulls attention exactly where you don't want it. A single fixed level, set correctly, sounds calmer.
In practice
Choose the bed for what it has to survive
Sparse music mixes far lower than busy music and still reads. A sustained pad or a single drone will sit 20 dB down and remain present, where a track with melody, percussion or a clear rhythm fights your voice for the same attention and has to be pushed further down before it stops interfering — at which point you can barely hear it and may as well not have used it.
Anything with a recognisable pulse also fixes the pace of the session, which is a problem when your pauses are meant to stretch. The music arrives at its next moment whether or not the listener is ready, and that quietly rushes the practice.
Scripts written for recording, with the pauses marked
Every silence timed and every segment noted for delivery, so you can set your bed against a known shape. Free to record and publish under your own name.