How to fix music that suddenly gets louder partway through
Tell Vizard Agent the music jumps in volume and roughly when. It reads the level second by second across the delivered file rather than trusting the file's average, finds the point where the track changes section, and rebuilds the mix with a bed that stays at one level from beginning to end.
What is the short version?
A file can measure perfectly on average and still make you jump halfway through. The average is one number for the whole video; the thing that startled you is a few seconds inside it, and only a reading over time shows it.
- Go to Vizard Agent with the video.
- Say the music jumps and roughly where.
- Ask for the level measured over time, not overall.
What do you need before you start?
The file that actually startled you and a rough position. "About halfway", "when the end card comes in", "the second half" are all precise enough — the exact second gets found by measurement, and your job is only to say that it happens.
- The delivered file. The one you downloaded.
- Roughly when it jumps. A half is fine.
- How much. "Noticeable" and "made me jump" differ.
- What the music is for. A bed, or a feature.
- Whether a music-free version helps. Sometimes it is the answer.
What do you type into Vizard Agent?
Say it jumps rather than that it is loud. Those are different faults with different fixes — "too loud" gets the whole bed turned down, which leaves the jump exactly where it was and makes the rest of the video too quiet.
Prompt
Variants worth knowing:
- "Until the music changes." Points at the cause.
- "Over time, not overall." The measurement that finds it.
- "One level the whole way through." The result you want.
What does Vizard Agent actually do?
Here is the order Vizard Agent worked in on a real set of short videos whose music jumped in the second half. The fix takes several attempts because the first few treat it as a mixing problem when it is a problem with the track.
- Measures the music's loudness when the bed is first built.
- Checks the loudness of the renders alongside the frames.
- Measures speech against music across the whole file.
- Reads the summary loudness of both sections separately.
- Measures the music's level over time rather than as one number.
- Reads its short-term levels to find where the jump sits.
- Smooths the transition into the section that jumped.
- Verifies the music no longer jumps at that point.
- Compares the level in the new file against the old one.
- Reads the level second by second in the delivered file.
- Rebuilds with every piece level-matched and no music at all.
- Puts the music back at one steady level on top of that.
- Checks the level is steady all the way through.
- Listens to the music track itself for section changes.
- Builds a uniform bed with no sections in it and re-renders.
Step fourteen is the actual diagnosis and it comes late. The mix was not wrong: the piece of music had a quiet opening and a loud second half, so every attempt to balance it against the speech balanced the average and left the change intact.
Step eleven is the technique that isolates the problem. Rebuilding the whole video level-matched and silent, then adding the bed back as one layer, means the only thing that can be uneven is the bed — which turns an argument about the mix into a single measurable thing.
What does the result look like?
A file whose music sits at the same level from the first second to the last, under speech that stays where it was. Nothing swells, nothing startles, and the average is the same as it was before — because the average was never the problem.
If the track will not behave, a bed without section changes replaces it.
When does this not work well?
Some music is meant to change as it goes. A track with a deliberate drop, a build or a key change carries its dynamics as the whole point, and having Vizard Agent flatten it removes exactly what made it worth using in the first place.
- Music with a deliberate build. Flattening it costs the effect.
- A feature track. If the music is the content, it should move.
- Very short videos. There is no room for a change anyway.
- Speech that varies wildly. The voice may be the uneven thing.
- A source that is already clipped. That is damage, not level.
How do you fix a result that came back wrong?
Say where it still jumps, or whether it now feels flat. Vizard Agent keeps the second-by-second readings and the beds it built, so a change is to one layer at one point rather than a fresh mix of the whole video.
- "It still jumps at the end." That transition smoothed and re-read.
- "Now the music is too quiet." The whole bed lifted, evenly.
- "The voice dips under it." Speech and bed re-measured against each other.
- "Just give me it without music." A music-free version built.
How does Vizard Agent compare to doing it yourself?
By hand this gets measured with one number for the file, which is what most tools report, and one number cannot show a jump. The instinctive fix — turning the music down — addresses the loud half and leaves the quiet half inaudible.
| By hand | Vizard Agent | |
|---|---|---|
| The measurement | One average for the file | Level read second by second |
| The diagnosis | "Too loud" | Where and why it changes |
| The fix | Turn the bed down | Even the bed, or replace it |
| The track itself | Not examined | Listened to for section changes |
| Proof | Playing it again | Levels re-read on the delivered file |
Common questions
Why did it measure fine? Because an average hides a jump. Vizard Agent reads it over time.
Can it just flatten the music? Yes. Vizard Agent can even the bed out or build a flat one.
What if the track is meant to build? Say so. Vizard Agent will keep the build and control the peak.
Can I get a version with no music? Yes. Ask Vizard Agent and it renders one.
Will the speech level change? Not unless you ask. Vizard Agent measures both separately.
Does this affect the whole video? Only where it needs to. Vizard Agent fixes the layer, not the file.
Can it show me the readings? Yes. Ask Vizard Agent for the level over time.
What about the platform's own normalisation? That is a separate step; Vizard Agent delivers to the target loudness.
Will it clip anywhere? Vizard Agent checks headroom and reports the peaks.
Can it match several videos to each other? Yes. Say so and Vizard Agent levels them as a set.
What if the jump is in my source clip? Same method. Vizard Agent measures the source too.
Does a fade fix it? Sometimes. Vizard Agent tries smoothing before replacing the track.
Can it pick a steadier track? Yes, and it will listen for section changes before using one.
Why measure over time rather than overall? Because you did not react to the file's average, you reacted to four seconds of it.