Duck music under VO to -18dB

If you cannot hear the claim, the art failed. Music is support, not a co-host. When the bed sits too hot under narration, viewers miss the sentence that justified the click. They do not write you a note about loudness. They leave.

A practical target many editors use is ducking music under voiceover to around -18 dB relative to the VO peak region, then adjusting by ear on phone speakers. The number is a starting line, not a religion. The principle is non-negotiable: speech intelligibility first. Hot mixes feel exciting in studio headphones and fall apart on a phone in a kitchen.

Why hot music destroys retention

Retention curves punish confusion. If a key phrase is masked, the brain works harder. Harder listening is still listening, until it is not. On mobile, with street noise and cheap earbuds, your "energetic" bed becomes mush. Creators romanticize a hot mix because it feels exciting at the desk. The audience is not at your desk. They are half-watching while cooking. Mix for that person.

Also, captions do not fully rescue a bad duck. Captions help, especially with sound off, but many viewers watch with sound on and still rely on the ear for emphasis. If the VO is fighting drums and a bright synth lead, the emotional punch of the claim gets flattened. You spent credits on a clear voice take, then buried it under a bed that wants to be the hero.

Compare music sources with the same discipline. Whether you use tracks from a comparison like Suno vs Udio vs Stable Audio, the mix problem stays human. A great generator does not excuse a bed that sits on top of speech. Instrumental beds with softer mids usually duck more cleanly than singer-forward demos. If your generator keeps handing you vocal-heavy previews, constrain the prompt or pick another take.

Watch section changes. A mix that works under calm explanation can fail under a dense list of steps. Numbers, names, and short commands need extra headroom. If the viewer misses step three because a snare crack sat on the word, the tutorial fails no matter how pretty the timeline looks. Treat ducking as part of the brief, not a finishing surprise. If the episode is number-heavy, plan more pauses where music can lift. If the episode is story-driven, plan a quieter bed overall.

How to duck without killing the track

Normalize or level the VO first. Get a consistent spoken bed.
Place music underneath at a low starting level.
Automate ducking under phrases, not under the whole timeline as one static smash.
Lift music in pauses, lower it when the claim lands.
Check on phone speaker and earbud. If either fails, the mix fails.

Sidechain-style ducking is useful when the music has a steady pulse. Manual rides are better when the VO has uneven energy, which most educational reads do. Do not be afraid of visible automation lanes. Pretty curves that protect consonants are professional, not fussy.

Watch the frequency fight, not only the loudness fight. A vocal-heavy music bed will clash even at a lower level. Prefer beds with less midrange clutter under spoken word. High-pass the music gently if your tools allow, so the VO owns the speech band. Low-pass harsh hats if they spit on top of S sounds.

For AI voiceovers, keep style lock consistent across a series so your ducking rides transfer. Style-locking AI text to speech in 2026 matters here because a voice that changes tone every episode forces remixing from scratch. Same voice, same target loudness, same duck depth becomes a house sound your audience trusts. Save a preset when you find a combination that survives phone checks. Future episodes should start from that preset, then adjust for denser scripts. Reinvention every Tuesday is how mixes drift and quality feels random.

Client feedback loops get easier when you can show two mixes: one with a hot bed and one with a disciplined duck. Most stakeholders pick the clear one when they hear both on a phone. The debate was never about taste. It was about default volume habits. When a client asks for "more energy," translate the request. Sometimes they want faster pacing or a brighter arrangement, not a louder bed under speech.

Captions, music, and the claim

Burned captions should match the spoken claim, not a paraphrase that drifts. After the mix is intelligible, run captions. An AI caption generator is fastest when the audio is clean enough for reliable transcription. Hot music can also hurt auto transcription, which then hurts your caption pass. Clean ducking is therefore an editing convenience, not only a viewer kindness.

Design caption timing to land with the duck. When music swells in a pause, keep captions sparse. When VO returns, captions and speech should agree. Viewers who watch with sound on still glance at text. Conflict between ear and eye feels sloppy. Avoid caption styles that cover the only visual proof on screen. Intelligibility includes seeing the object the VO names.

If you publish with optional burned-in captions and separate platform captions, keep the wording aligned. Dual systems that disagree make the channel feel careless. Carelessness is sticky in the wrong way. Spot-check numbers and proper nouns at 1x speed after captions land. A wrong digit in a caption next to a muffled VO is a trust tax.

A repeatable mix checklist

VO solo: every claim clear.
Music solo: energy matches section, not louder than needed.
Combined on phone speaker: claims still clear.
Combined with light room noise or a fan in the background: still clear enough.
Final: music rises only when speech rests.
Spot-check numbers and proper nouns at 1x speed.

If you produce faceless series at volume, save a mix preset: VO target, music duck depth, limiter ceiling. Reuse it. Consistency trains your ear and reduces "why does this episode feel worse" mysteries. Do not chase loudness wars for YouTube. Platforms normalize. Clarity travels better than brick-wall excitement.

Producers who wait until export day to think about music end up choosing the loudest track that "feels done" and then fighting it for an hour. Pick the bed earlier, at a safe duck depth, while you still have room to change arrangement. When you want generation, voice, and caption tools in one place so mix decisions happen earlier, sign up at Versely and keep the audio bed as part of the same project habit, not a panicked last step. The channels that sound calm and confident usually ducked harder than their first instinct suggested.

FAQ

Is -18 dB a hard rule?

No. It is a useful starting duck depth under VO. Your genre, voice, and music arrangement may need more or less. Phone checks decide.

Should music be silent under every sentence?

Not necessarily. A low bed can keep pace. It should never compete with consonants and numbers. If viewers miss a price, a step count, or a name, duck harder.

Can I fix a hot mix with louder VO alone?

Sometimes, briefly. Loud VO over loud music still fatigues. Better to lower the bed and keep speech natural than to push both into a harsh ceiling.