Mixing Voiceover With Music: Levels That Actually Work

Everyone starts with the same fix: the music is covering the voice, so turn the music down. It works for a while. Then it stops working — the music is now so quiet that it contributes nothing, and the voice still sounds slightly buried rather than clearly in front.

That happens because volume was never the whole problem. A voice and a piece of music occupy overlapping frequency ranges, and when two sounds share a range, the louder one masks the quieter one in that band regardless of the overall balance. This guide covers the level relationship to aim for, how to let the mix handle itself, and what to do when level has run out of road.

Set the Level While Someone Is Talking

The most common mixing error takes ten seconds to make. You drop the music in, listen to the opening — which is instrumental — set a level that sounds good, and move on.

That level was set under the only condition where the music has no competition. As soon as the voice arrives, the same setting is too loud, and you have now built the whole mix on a judgement made in the one place it does not apply.

Set the music fader with the voice playing over it. Find a section where the speaker is in normal delivery, not their loudest or quietest, and balance there. Then check the passage where the music is busiest, because that is where the relationship will fail first.

The target is a relationship rather than a number: the voice unambiguously in the foreground, the music clearly present and not merely audible. If you can hear the music but could not describe it, it is too low and it is doing no work. If you find yourself concentrating to follow the words, it is too high.

Let Ducking Do the Routine Work

Riding the music fader by hand for a ten-minute video is a great deal of work with an uneven result. Automatic ducking does the routine version of it.

The mechanism is a compressor on the music track keyed by the voice track. When the voice is present, the music is pulled down by a set amount; when the voice stops, it returns. The settings that matter are how far it drops, how quickly it reacts, and how quickly it recovers.

A fast attack catches the start of each phrase, which is what you want — a slow one lets the first word get covered. Release is the setting people get wrong: too fast and the music surges audibly between every sentence, which is more distracting than the original problem. A slower release keeps the music down through the natural gaps in speech and only lets it back up when the speaker has genuinely finished.

Ducking handles the average. It does not know that the third paragraph is the important one, or that the music should swell under the closing line. Those are automation, done by hand, on the handful of moments that carry weight.

Carve Space Instead of Turning Down

When level and ducking are both set sensibly and the voice still is not clear, the problem is frequency.

Speech intelligibility lives largely in the presence region — the upper midrange, where consonants are defined. Most music also has content there: guitars, synth pads, the top of a piano, cymbals. Both sounds are competing for the same part of the spectrum, and the ear cannot separate them by volume alone.

The fix is to make room. A gentle broad cut in the music around the presence region, only a few decibels wide and shallow, opens a gap the voice sits in. Done subtly, nobody notices the music has been altered; they notice the voice got clearer.

Two supporting moves help.

High-pass the music. Roll off the very low frequencies in the music track. Voice has almost nothing down there, so you lose nothing from the speech, and you remove low-end energy that muddies the mix and eats headroom.

High-pass the voice too, lightly. Rumble, handling noise and room resonance sit below the useful range of speech, and removing them makes the voice sound tighter without changing its character. This belongs in the dialogue chain rather than the mix, and it is the first step of the sequence in the order to clean up bad audio.

Where the two still fight after all of this, the music is the wrong music. A track built around a strong midrange lead will never sit comfortably under speech, and choosing a sparser one is faster than any amount of processing.

Choose Music That Can Take a Voice

Selection does more for a voiceover mix than processing does, and it is free.

Sparse beats dense. A track with a few sustained elements leaves room; one with a constant busy arrangement does not. Instrumental beats vocal, without exception — two voices in the same range compete directly and no processing separates them cleanly.

Steady beats dynamic. Music with dramatic swells is fighting your speech for emphasis, and the swells never land where the words need them. Save dynamic tracks for passages with no narration.

And watch the length. A track that has to loop for eight minutes becomes noticeable in a way it never is on a first listen, so either choose something long enough or plan where it drops out. Silence under a section is a legitimate choice and often a stronger one than music that has outstayed its purpose.

Levels at the End

Once the balance is right, the overall level is a separate decision, and it is the last one. Do not raise the master to make the mix feel louder — that changes nothing about the voice-to-music relationship and only pushes the whole thing toward a limiter.

Delivery level is governed by what the destination does with it, and most platforms will adjust your file to their own target whatever you send. Aiming at that target rather than above it means the mix arrives with its dynamics intact, which matters more for speech than for almost anything else. The measurements and the reasoning are in audio loudness standards and why your video sounds quiet.

A related check: the mix should still work in mono. Some playback situations sum the channels, and elements placed wide in stereo can partly cancel when that happens. Switch to mono once before delivering and confirm nothing important disappears.

Next Steps

Open your last video with music and do one test: solo the music, play a minute with the ducking active, and listen for the track being pulled around. Most mixes have a release time set too fast, and slowing it down is a thirty-second change that removes a restlessness nobody had identified.

Then check whether your music was chosen or merely found. A sparser instrumental track will make every subsequent decision easier than any processing applied to a busy one. Where the voice itself is the weak part of the chain, the fix is at capture — recording a clean reference clip covers what that takes, and the rest of the voice material is under AI voice and TTS.