8. Giving the Silent Scene Its Sound

A soundtrack built in layers

The courier scene is locked. Nine shots sit on the Shotcut timeline, and the only sound under them is the ambience bed. The picture won't change now. That matters for sound, because every effect you place will be tied to a frame. If the picture moved, every effect would slip out of sync with it.

This chapter answers one question. How do you give a silent scene a soundtrack that makes it feel real? Picture tells you what happened. Sound tells you it was real, and where you are. A knock you can see but can't hear looks like a mime. Rain you can see on a jacket but can't hear makes the street feel like a stage set. So each sound in this pass goes in for a reason. It fills a gap the viewer would notice if the sound were missing.

The walkthrough is illustrative. The sounds, levels and file choices are examples of how the work goes. They are not a record of anything you have done.

A soundtrack like this has layers. The bottom layer is ambience. That is the steady background sound of a place, such as rain on a street or traffic far away. On top of it sit spot effects. These are single sounds tied to single actions, such as a knuckle on a door or a parcel on stone. In film work, finding where those sounds belong is called spotting. Then comes the mix, which is the balance between the layers. Last comes the check: you listen, and you read the meters.

Watch it with the sound off

Mute the timeline and watch the nine shots straight through. Then watch again and stop at every moment where something visible should make a noise. Write each one down. This is your spotting list. It works like the shot list from chapter five. There is one row per sound, with the shot number, the action, the sound you want and roughly where in the shot it falls.

Say your list comes out like this. In the knock shot, two knocks on the bakery door. In the crouch shot, the scuff of a boot on the wet step, then the parcel set down. In the phone close-up, a small tap on the screen. Through every outdoor shot, rain on the jacket. In the wide walk, footsteps heading toward screen right. In the arrival at the lit door, the footsteps slow and stop.

Watching with the sound off forces you to really see. With the ambience playing, your ears fill in the gaps, and you stop noticing that the parcel lands without a sound. With no sound at all, every silent action stands out.

Before you go looking for sounds, check one thing in the list. Does every row match something you can see? Leave out a sound for something offscreen unless it does a job. A dog barking that nobody reacts to only gives the viewer a puzzle.

Getting the sounds, and checking the licence on each

You can get sounds in two ways. You can download recorded ones from a sound library, or you can generate them from a text description. Either way, each file needs a licence check before it goes on the timeline. This carries on the asset licensing from chapter two. A video can be blocked or demonetized because of one bad file, even if everything else in it is fine.

The library to start with is Freesound. It is a large free collection of sounds uploaded by its users. You make a free account, search for a sound such as "knock wooden door", and filter by licence. On a sound's page you click Download. Most files come as WAV, FLAC or MP3. WAV is the best choice for editing, because it hasn't been compressed.

The licence filter is the step that matters. Freesound sounds carry different Creative Commons licences, and for a YouTube upload they behave very differently. CC0 is public domain. You can use it in a monetized video with no credit needed. CC BY is also allowed in a monetized video, but you must credit it. That means naming the creator, linking to the sound and to the licence, and saying whether you changed the sound. CC BY-NC means non-commercial only. It is not allowed on a monetized channel. So if you plan to earn from the upload, filter to CC0 and CC BY, and keep the credit details for every CC BY file.

There is a simpler option inside YouTube itself. YouTube Studio has an Audio Library with a Sound effects tab. Those effects are cleared for monetized use in YouTube videos, with no credit needed. The limit is that they're cleared for use inside YouTube videos. You can't repackage or resell them as sound files. For a scene that will only ever live on YouTube, this is the low-risk choice. The files come as MP3, though, which is fine for short effects.

Then there's generation. ElevenLabs has a Text to Sound Effects tool. In the ElevenLabs dashboard you open Sound Effects and type a description. You set a duration between one and eleven seconds, click Generate Sound Effect, listen to the result and download it. A description in the spirit of "two firm knuckle knocks on a wooden shop door, rain in the background, close" gives you something to judge.

Here the licence depends on your plan. On the free plan, output is non-commercial only and needs credit. So it isn't allowed in a monetized video. Paid plans, which start at Starter, include a commercial licence. That covers sounds used inside a larger work such as a YouTube video, with no credit needed. Use the free plan to test whether the tool can make the sound you want. If you keep the result for a monetized release, make the final version on a paid plan.

When should you generate instead of searching? The knock is the case for it. A library has thousands of knocks, but hardly any will match two knocks with the exact gap the courier's hand makes in the picture. A generated knock can be described with the rhythm you need. Footsteps on wet stone and steady rain are easy to find in a library. Recorded sounds also tend to sound more natural over a long stretch. A good split is library for the ambience and the common effects, and generation for the one or two sounds that have to match a specific gesture.

Whatever you generate, treat it like a generated shot. Listen to what actually came out. Don't assume the prompt describes it. Generated knocks sometimes come back as three knocks, or with a strange metallic ring. Reject those takes and log them, the same way you log rejected video takes.

For each file, add a line to the production records. Write the file name, where it came from, its licence, any credit owed and which shot it serves. For a CC BY sound, copy the credit details then and there. Tracking down an author weeks later is tedious, and sometimes it can't be done.

Laying the tracks out in Shotcut

The ambience bed already has its own audio track. Spot effects need a separate one. In the Timeline panel, open the Timeline menu, which is the three-line icon above the track headers. Choose Track Operations, then Add Audio Track. The shortcut is Control U, or Command U on a Mac, the same one you used for the animatic. Add a second effects track as well. Then two sounds that overlap, such as the footsteps and the rain on the jacket, can each sit on their own track.

Keeping the layers on separate tracks lets you adjust one without touching the others. If the knocks are too loud, you change the effects track and the rain stays as it was.

Next, make sure you can see each sound's shape. In the Timeline menu, check that Options, then Show Audio Waveforms, is turned on. A waveform is the drawing of a sound's loudness over time. Quiet parts are thin and loud parts are tall. If a newly imported file shows no waveform, right-click the track header and choose Rebuild Audio Waveform.

Now sync the first knock. Move the playhead to the frame where the courier's knuckle meets the door. Use the left and right arrow keys, which move one frame at a time. Find the frame where her hand stops moving forward. That is the contact frame. Turn on snapping with Alt P, or click the magnet icon. Zoom in with the plus key or Control and the mouse wheel. Look at the knock file's waveform. A knock starts with a sudden tall spike called the transient. That spike is the moment of impact. Drag the clip until the spike snaps to the playhead.

The second knock is the real test. If your file holds both knocks, the gap between them in the sound has to match the gap in the picture. Step to the second contact frame and see whether the second spike lands on it. If it's a few frames early, split the file between the knocks and slide the second one along. If the generated knock got the rhythm wrong, you now have a concrete reason to make a new take.

Each frame at the project's frame rate is a fraction of a second, and people notice when sound and picture drift apart. A sound that lands a frame or two after its action usually still reads as connected. A sound that lands before its action looks wrong sooner. So if you can't get it exact, it's safer to land the sound a hair late than early.

Do the same for every row on the spotting list. Put the parcel's soft thud on the frame where it touches the step. Line up the footsteps with the frames where each heel strikes the ground. Here the waveform helps a lot, because a run of footsteps shows up as a row of spikes. You can line them up with the steps you see, one by one.

Carrying sound across the cuts

Picture cuts are instant, and sound that cuts with the picture sounds broken. Imagine the rain stopping for a split second at every cut. The viewer would feel the edit. The fix is to let the ambience run under the cuts as one continuous clip, or as clips that overlap, rather than cutting it with each shot. The street's sound keeps going while the camera angle changes, just as it would if you were standing there.

Cuts can also cause a different problem: a click. If an audio clip starts or stops in the middle of its waveform, the sudden jump makes a small pop. Short fades fix this. Hover over the top-left or top-right corner of an audio clip until a round handle appears. Then drag inward to draw a fade-in or fade-out. A few frames is enough to remove a click on an effect. For more control, select the clip, open the Filters panel, click the plus button and choose Fade In Audio or Fade Out Audio from the Audio tab. These let you set the length and the shape of the curve. The shapes include Natural and S-curve.

There is one place where the ambience should change, and that's where the location changes. The bakery step is a dark shopfront in the rain. The lit door above the shop next door is a different spot. It is slightly sheltered, and maybe there's a faint sound of a home behind it. When the courier walks off toward screen right and arrives at the lit door, the sound can tell the viewer she has moved somewhere new.

To do that, overlap two ambience clips across the cut, the street rain and a softer, more sheltered rain. Fade the first out and the second in over about a second, so neither one stops dead. This is called a crossfade. Start it a little before the picture cut and you have an old editor's trick. The sound arrives first and pulls the viewer into the next place.

Balancing the layers

With everything placed, play the scene and listen to the balance. Usually something is wrong. Effects straight from a library tend to be too loud. They were recorded to be heard on their own.

Shotcut's Gain/Volume filter controls how loud a clip is. Select a clip, add Gain/Volume from the Audio filters and set its Level in decibels. Decibels measure loudness on a scale where each step is a ratio, not a fixed amount. A drop of about six decibels sounds clearly quieter without disappearing.

Sometimes a level needs to change partway through a clip, and keyframes handle that. Next to the Level slider is a stopwatch icon that opens the Keyframes panel. Set a keyframe where the change should start, step forward and set another at the new level. Shotcut moves smoothly between them. Right-click a keyframe to choose how it moves: Linear, Smooth or Hold.

In this scene, keyframes do a quiet job. When the courier reads the sign and the shot tightens to her reaction, dip the rain a few decibels. The moment then feels more inward. Bring the rain back up as she picks up the parcel and moves. The viewer won't notice the dip itself. They'll only feel the moment go still.

The opposite mistake is effects that bury the ambience. If the footsteps drown out the rain, the street goes away and she seems to be walking in an empty room. Lower the footsteps until the rain sits under them the whole time.

Now the meters. Open View, then Scopes, then Audio Peak Meter, or switch to the Audio layout. The peak meter shows the loudest moment at each instant, in decibels below full scale. Zero is the top of the scale. Anything that reaches zero clips. The waveform's top gets cut off, and you hear a harsh crunch. Keep the loudest effects, here the knocks, a decibel or two below zero, so you have some room to spare.

The peak meter tells you how loud the loudest instant is. It doesn't tell you how loud the whole scene feels. For that, open View, then Scopes, then Audio Loudness. It measures loudness in LUFS, which stands for loudness units relative to full scale. That is a measurement built to follow how loud something sounds to a person. The Integrated reading is the loudness of the whole piece. True Peak is a stricter version of the peak reading.

YouTube adjusts uploads toward an integrated loudness of minus fourteen LUFS, with true peaks kept at or below minus one. The adjustment only goes one way. A video louder than minus fourteen gets turned down. A quieter one is left as it is and is not turned up. So a quiet mix plays quiet next to everything else, and a mix pushed loud gets turned down anyway, and pushing it loud may have cost you clean sound.

To hit the target, use the Normalize: Two Pass filter. Add it, enter minus fourteen as the target, click Analyze and let Shotcut scan the clip. It then applies one fixed gain change. The relative levels you balanced stay exactly as they were. There's also Normalize: One Pass, which adjusts as it goes. Set badly, it can make the level audibly pump up and down. For a short finished scene, Two Pass is the safer choice. Afterward, play the scene through with the loudness scope open and check the integrated figure and the true peak.

Listening twice

Meters catch clipping and loudness. They can't tell you if the scene feels real, so listen to it twice in two different ways.

First, listen on headphones. They show you detail: a click at a cut, a knock a frame late, a gap where the ambience drops out. Then listen on laptop speakers. They are small and thin, and they show you balance. A low thud that sounded full on headphones may vanish. Rain that sounded gentle may hiss. A mix that works on both will work on most of what your viewers use.

On both listens, check the same four things. Is anything crunching, which means clipping? Are there gaps or clicks at cuts? Does any sound land off its action? Is any effect burying the ambience? When something fails, trace it the way you traced faults in the picture: find the one input that caused it and change only that. If one knock is late, move that clip. Don't redo the whole effects track.

Export and keep the records

Export the scene again with the YouTube settings you've used since chapter two. Then play the exported file itself, because that is what the viewer will get. Listen to it once. Confirm the sound survived the export and is still in sync.

Save the project file, the sound files and the records together. The record for each sound covers where it came from, its licence and any credit owed. If a claim ever arrives months after upload, those records are how you answer it. The scene now has a clean picture and a sound pass. Titles and a consistent grade are the next steps toward the finished short.

A Foley tool that starts from your video

One release this month is aimed at exactly the job above. On October twentieth, 2026, NC AI released VARCO Sound. It is a web platform for building generated soundtracks. It takes text, an image or video as input and generates synced effects and background audio. It works in multiple tracks and arranges them on a timeline you can edit. There is a chat-style assistant called the Sound Agent that arranges the sound for you, plus a desktop app.

The production task it affects is spotting and syncing. In this chapter you watched the cut, wrote a spotting list, found or generated each sound and lined each one up by hand. A tool that starts from video claims it can suggest those sounds and their timing from the picture itself. It can't take over the judgment, though. The tool doesn't know the courier's knocks must have a particular rhythm, or that the rain should dip at the reaction close-up. Treat what it makes the way you treat a generated shot. Check it against the frames, don't trust it.

The smallest useful next action is a one-shot test. Export the knock shot by itself. Run it through VARCO Sound. Compare where its knocks land with the contact frames you found by stepping in Shotcut. If it lands both knocks within a frame or so, it could save real time on longer scenes. If it doesn't, you've spent one shot finding that out. VARCO Sound is a commercial subscription. A launch offer runs to November thirtieth with half off the first month. Before using any of its output in a monetized upload, check the licence terms on the tier you're on, the same check you ran on every sound in this chapter.