How to Automatically Lower Background Music Under Dialogue with AI

The music feels right over the food close-up. Then the cook starts explaining the next step, and you have to strain to hear it. Turning the whole track down fixes the speech but drains the energy from the shots between instructions.

Audio ducking changes the music level around the voice: lower during speech, higher when there is room for it.

Doing this by hand means finding each spoken passage, lowering the music around it, and shaping every rise and dip so the changes feel natural.

In invideo Editor, you can give that job to the built-in AI agent, then develop the mix through further instructions or the professional timeline and audio controls.

The useful brief goes beyond “make the music quieter.” Tell the agent where music can lead, how it should return, and what must stay untouched. Then listen across those transitions.

Start with separate dialogue and music

Open your existing edit in invideo Editor and add the music track you want to use. If you started in Premiere Pro, DaVinci Resolve or another editor export a XML project with its source media to continue working on an editable timeline. Make sure your audio and music are on separate audio tracks so you can adjust the music without changing the voice.

For this walkthrough, imagine a 90-second lemon-pasta explainer for home cooks. The picture edit is already approved. It has three spoken passages, separated by a pan close-up and a plating close-up. A supplied instrumental should support the instructions and come forward during those two breaks.

Give the agent context first:

Work on the Lemon pasta timeline, the approved 90-second cooking explainer. Use my supplied Kitchen instrumental.wav beneath the existing dialogue. Keep the music separately editable. Preserve the picture edit, every spoken word and the current dialogue timing. Use only my supplied assets; do not generate or source new music.

This establishes the material and the boundaries. The next instruction sets the relationship between voice and music. If you still need a track, you can add music in invideo Editor by choosing stock music or generating a new track. This example uses an uploaded instrumental throughout.

Lower the music around the spoken passages

Ask the built-in agent to identify speech in the edit rather than making a list of every sentence yourself. Describe the listening result and distinguish a brief breath from a deliberate break in the explanation.

Lower Kitchen instrumental.wav whenever the cook speaks, so the instructions stay easy to follow. Keep the music low through short pauses within a sentence. Let it rise smoothly during the two food close-ups with no speech, then settle beneath the voice before the next instruction begins. Give the ending a gentle fade. Preserve the dialogue level, all spoken words and the entire picture edit.

Lowering the whole music track gives you one fixed balance; ducking changes that balance around speech.

You can keep refining the balance between dialogue and music after the initial request, using further instructions or the editable audio controls.

There is no universal music level that makes every voice clear. A sparse instrumental, a busy arrangement and a softly recorded speaker need different balances. Begin with understandable dialogue, then bring in enough music to support the scene.

Listen to the handovers, not just the loudest moment

Play from before each spoken passage into the following close-up. Then play from that close-up into the next line. You need to hear the lead-in and the recovery together.

• The first word: Music should already leave room when the cook begins speaking. A dip that arrives after the first syllable is too late.

• A sentence pause: A breath should not trigger a conspicuous swell followed immediately by another dip. Keep related phrases feeling like one explanation.

• The food close-up: With no voice to carry the sequence, the music can become more present. Choose a rise that supports the shot rather than making the listener reach for the volume.

• The returning voice: Hear the final musical beat and the next instruction together. The voice needs room before the new line begins.

• The ending: Listen through the final word and the music tail. A fade should resolve the sequence without cutting the last instruction short.

Use the waveform and volume changes to locate a passage, then judge it in playback. A convincing curve on the timeline cannot tell you whether a quiet word remains audible.

Refine one recovery while keeping the approved mix

Suppose the second close-up, around 01:02–01:13 in this example, has a satisfying musical rise. But the recovery stays too loud as the cook says “save a little pasta water.” Lowering the entire track would also flatten the first close-up, which already works.

Give a bounded revision:

The timing identifies the passage; the quoted line identifies the important boundary. Adjust both to match your actual edit.

Replay from before the rise through the whole returning sentence. If the music still draws attention to itself, ask for a smaller rise in that interval. If the transition sounds smooth and the line is clear, preserve it and move on.

Shape the finish manually when you know what you want

You can do the audio work directly in the same editor. The agent is useful for the repeated speech-and-music pass; manual control is useful when you want to shape one entrance or compare a very small change immediately.

Use the audio editing workspace to select the music and adjust its level. Its current guidance also describes a Ducker effect for refining how far music drops and how it returns. You can direct the agent or work with that effect yourself.

For a local change, refine the music’s volume keyframes or fades around that passage, leaving the dialogue timing in place. A track-level fader changes the overall balance; timed volume changes shape the rise and fall within the sequence.

Solo the dialogue to check the line itself, then restore the music and listen in context. Mute the music briefly if you need to identify whether the problem comes from the recording or the balance. Finish with all intended layers playing. The editable sound-bed guidance also covers moving and trimming sounds, track volumes, pan, solo and mute.

In a separate built-in-agent sound-design example, Joseph opens the audio clip’s volume, pan and fade controls. That is a useful view of manual finishing in the Editor; it is a sound-effect adjustment, not an audible ducking comparison.

If you also want atmosphere or action accents around the food shots, use the guide to adding sound effects and music. Balance the voice and music first so those extra layers have a clear place in the mix.

Check the whole video and export

Listen through all 90 seconds once the local revisions work. A change that sounds right in isolation may feel too dramatic beside the rest of the music. Check the mix on headphones and ordinary speakers, at a comfortable playback level.

Export the finished video, then listen to that file as well. Confirm complete words, both close-up recoveries and the final fade. For the rest of the editing process, the first-cut-to-export walkthrough shows how sound fits alongside picture, colour, captions and finishing.

Manual editing and video exports are free, without a watermark. Work by the built-in AI agent uses credits, and generating new media uses credits too. The amount varies with the assignment and models. You can inspect usage for an exchange and the project total; see the current Editor credit guidance. Using a supplied track avoids requesting new music generation, but the agent’s editing work still uses credits.

Frequently asked questions

Is audio ducking the same as turning down background music?

Turning down the whole track sets a quieter overall level. Ducking lowers music around speech and lets it recover elsewhere. Use it when the voice needs priority but music should still carry the gaps.

How low should background music be under dialogue?

Low enough that the words are easy to understand at normal playback volume. Choose the balance by listening to your actual voice and track together; a single dB value will not suit every recording.

Can I duck music that is already mixed into my video?

If voice and music are baked into one audio file, lowering that file lowers both. Separate dialogue and music sources give you independent control. Putting a mixed recording on its own track does not separate its contents.

Can I change the AI’s music adjustments manually?

Yes. Work on the editable timeline and audio controls to refine levels, volume changes and fades, or ask the agent for a focused revision. Specify which approved voice and picture elements must stay unchanged.

Does automatic music ducking use credits?

Built-in-agent work uses credits. Manual audio adjustments and video exports are free. Generating a new music asset also uses credits, separately from directing an edit with your supplied track.