Mixing Music Under Voiceover So Words Stay Clear

Quick answer: Keep background music 15 to 25 dB below the voice while someone is speaking, and let it rise in pauses. Use ducking (a sidechain compressor or your editor's auto-duck) with about 6 to 10 dB of reduction, a 10 to 30 ms attack and a 300 to 600 ms release. Cut 2 to 4 dB around 1 to 4 kHz in the music so consonants cut through, and pick instrumentals with no lead melody. Master the finished video to about minus 14 LUFS for YouTube.

This guide is for anyone placing music under speech: YouTubers recording voiceovers, explainer and ad editors, course creators, podcasters with narrated segments. It assumes you can move a volume fader and add an effect in your editor, nothing more.

You will get concrete numbers for levels, ducking and EQ, a before and after mix walk-through, and the creative side most mixing guides skip: choosing or generating royalty-free background music that leaves room for a voice before you touch a single fader. Good music choice solves half the problem.

Why words disappear under music

Speech intelligibility lives mostly in the consonants: the "t," "s," "k" and "p" sounds that carry most of the information in a sentence. Those sit roughly between 1 and 4 kHz, with sibilance higher. Vowels, which carry loudness and warmth, sit lower.

Music masks speech in two ways:

• Level masking: the music is simply too loud overall.
• Frequency masking: the music has energy in the same range as consonants, so even at a modest level it blurs them.

Most people fix the first and forget the second. They pull the music down until words are clear, and the bed ends up so quiet it might as well not be there. Fixing frequency masking lets the music stay audible while the voice stays clear.

Target levels you can actually measure

"Make the music quieter" is not a level. Here are numbers you can set on a meter.

Measure both on the same type of meter. If your voice averages around minus 16 LUFS short-term, a bed at minus 36 to minus 40 LUFS short-term is in the 20 to 25 dB range. If you only have peak meters, aim for voice peaks around minus 6 to minus 3 dBFS and music peaks around minus 24 to minus 20 dBFS under speech.

Then master the whole program. For YouTube and most video platforms, an integrated loudness around minus 14 LUFS with true peaks below about minus 1 dBTP is a common target. Podcasts often aim for about minus 16 LUFS stereo. Platforms normalise playback, so mastering hotter usually just gets turned down.

Ducking: settings that do not pump

Ducking lowers the music when the voice is present. Done well, nobody notices it. Done badly, the music breathes in and out between every word, which is called pumping.

Option 1: sidechain compressor

Put a compressor on the music track and set its sidechain input to the voice track. When the voice gets loud enough, the compressor turns the music down.

• Threshold: set so the voice triggers it reliably but room noise does not.
• Ratio: 3:1 to 6:1.
• Gain reduction: 6 to 10 dB on top of your static level difference.
• Attack: 10 to 30 ms. Faster clamps down on the first syllable; slower lets a click of music through.
• Release: 300 to 600 ms. Shorter pumps between words; longer stays down through short pauses, which is what you want.
• Lookahead: if your plugin has it, 5 to 10 ms helps the duck start just before the first word.

Option 2: your editor's auto-duck

Most editors include an automatic ducking feature that analyses the dialogue and writes volume keyframes on the music. These are fast and usually good enough for talking-head videos. Set the reduction amount, then check the fades in and out of each phrase, and smooth any that feel abrupt.

Option 3: manual volume automation

For ads and trailers, draw the volume by hand. It takes longer but gives full control, such as letting the music swell into a two-second gap before a tagline.

EQ: make a hole for the voice

A small cut in the music where speech lives buys more clarity than several decibels of fader.

1. Cut the music's presence range

   Add an EQ to the music bus with a broad cut of 2 to 4 dB centred around 2 to 3 kHz, with a wide Q. This leaves space for consonants.

2. Tame low mids if the voice is muddy

   If speech sounds thick, a gentle cut of 1 to 3 dB around 250 to 400 Hz in the music helps a deep voice.

3. Optionally make it dynamic

   A dynamic EQ band, sidechained from the voice, cuts only while the voice speaks, so the music sounds full in gaps.

4. Do not overdo it

   More than about 6 dB of cut makes the music sound hollow. If you need more, the track is the wrong choice for this voice.

Choose music that leaves room in the first place

Mixing fixes the edges. Arrangement decides whether a track can sit under a voice at all. The best voiceover beds share a few traits:

• No vocals, including chopped vocal samples, which compete directly with speech.
• No lead melody, or a very sparse one that stays out of the 1 to 4 kHz range.
• Pads, sustained chords and bass carrying the harmony.
• Light, steady percussion with few fills or builds.
• A stable energy level, so the music does not build to a drop in the middle of a sentence.

If you generate the bed, ask for those traits directly. Here are two prompts that consistently leave room for speech.

MusicVivo starts vocals at none for this kind of request, and you can set energy low and generate several variations to compare. For broader prompt technique, see the text-to-music prompting guide, and for instrumental-specific advice, the AI instrumental music guide.

Worked example: rescuing a muddy explainer mix

A 2-minute app explainer has a clear female voiceover and a bright pop track with a plucky synth lead. Viewers say it is hard to follow.

Before:

• Voice averages minus 16 LUFS short-term.
• Music sits at minus 26 LUFS short-term, only 10 dB below.
• The synth lead plays constantly around 1.5 to 3 kHz.
• No ducking, no EQ.

Fix, in order:

1. Replace the bed with a sparse version: same genre, pads instead of the synth lead. This alone fixes most of the problem.
2. Set the bed to about minus 36 LUFS short-term under speech, a 20 dB difference.
3. Add a sidechain compressor: 4:1 ratio, 20 ms attack, 450 ms release, 8 dB of reduction.
4. Cut 3 dB at 2.5 kHz on the music with a wide Q.
5. Let the music rise to about minus 20 LUFS short-term in the 4-second gap before the call to action.
6. Master to minus 14 LUFS integrated, true peak minus 1 dBTP.

After: every word is clear on a phone speaker, and the music still gives the video energy. The swell before the call to action now feels deliberate.

For the creative side of picking that replacement track, our guide on how to pick background music for videos covers tempo and energy. For podcast-specific levels, see background music for podcasts.

Adjusting for different kinds of voice

The numbers above assume one clear narrator at a steady pace. Real projects vary, and each variation pushes the settings in a predictable direction.

Fast ad reads. A voice delivering a 30-second script at speed leaves almost no gaps, so ducking barely releases. Rely on the static level difference instead, around 12 to 15 dB, and choose music with a strong rhythm but a simple harmonic bed. The energy comes from the drums, not from melody.

Soft or breathy voices. Meditation guides, ASMR-style narration and intimate storytelling have less consonant energy. Pull the music further down, toward 25 dB below, and favour very low or very high pads that leave the whole midrange to the voice.

Deep voices. A low male narrator overlaps heavily with bass and low-mid instruments. A gentle cut around 200 to 400 Hz in the music, plus a high-pass filter on the music around 60 to 80 Hz, keeps the voice from sounding boomy.

Interviews and multiple speakers. Different microphones and voices make one ducking setting unreliable. Drive the sidechain from a bus that carries all dialogue tracks, and consider leaving music out of the conversation entirely, using it only for transitions.

Translated or dubbed versions. A dubbed voice often has different pacing and pauses from the original. Recheck the ducking and any hand-drawn swells for each language version rather than copying automation across.

Voice recorded in a noisy room. Music is often used to hide room noise, but a louder bed only adds a second problem. Clean the dialogue with noise reduction first, then set the bed by the normal rules.

Check on the speakers your audience uses

Studio headphones lie in your favour. Before you export, check on:

If any device makes a word hard to catch, pull the music down 2 to 3 dB under that line or deepen the EQ cut slightly. Do these checks at the end of the day with fresh ears too, since hours of mixing make you tolerate a louder bed than a new listener will. Small speakers reproduce midrange and lose bass, so a bed that seems quiet on headphones can sit right on top of the voice on a phone.

Licensing the music you mix

Whatever you mix, it has to be licensed for where you publish. A track that is perfect under your voice but unlicensed for a monetized channel or a client ad is still unusable. MusicVivo paid plans include a commercial license covering monetized YouTube, TikTok, podcasts, ads and client work; the Free plan is personal and non-monetized. Read the full license terms, and see how to avoid Content ID claims for the records to keep. If you make voiced ads with jingles, our guide to choosing vocals in AI songs covers when a sung line helps.

People also ask

How loud should background music be under a voiceover? Start with the music 15 to 25 dB below the voice. Use the quieter end for explainers and podcasts and allow around 12 to 15 dB for energetic ads.

What is audio ducking? Ducking automatically lowers music while someone speaks and lets it rise in gaps. It is usually a sidechain compressor or an editor's auto-duck feature.

What ducking settings should I use? Try 6 to 10 dB of reduction, 10 to 30 ms attack and 300 to 600 ms release at 3:1 to 6:1. Lengthen the release if the music pumps.

Why is my voiceover hard to understand even when the music is quiet? The music is probably masking consonants between about 1 and 4 kHz. Cut that range in the music or choose a sparser track.

Related reading

• Corporate Background Music That Doesn't Sound Cheap
• Background Music for Ads and Social Campaigns
• Copyright-Free Music for YouTube: A Creator's Guide
• Background Music for Twitch Streams
• AI Music File Formats
• BPM and Key in AI Music

Product and company names mentioned in this article are trademarks of their respective owners. MusicVivo is an independent product and is not affiliated with, sponsored or endorsed by any of them. Comparisons are based on publicly available information at the time of writing; check each provider's own site for current features, pricing and terms.