Vizard Agent

How to raise one quiet section without the limiter shifting your audio

Last updated 2026-09-15 · 8 min read

Tell Vizard Agent which section is quiet and that the rest must stay untouched. It raises that stretch, limits only the peak the boost creates rather than the whole programme, then checks whether the limiter introduced a delay — and compensates for it, because a few milliseconds of shift puts the whole track out of sync with the picture.

What is the short version?

One section of dialogue is quieter than the others, so you raise it. That creates a peak, the peak needs limiting, and the limiter you reached for may quietly delay everything downstream of it — which turns a twelve-second level fix into a sync problem across the whole film.

  1. Tell Vizard Agent which section is quiet.
  2. Say the rest of the mix must not change.
  3. Ask it to check the audio did not shift.

What do you need before you start?

The film and a sense of which section sounds wrong. Vizard Agent measures rather than guesses, so a rough pointer — "she sounds quieter in the middle" — is enough to start, and the measurement decides how much correction is actually needed.

What do you type into Vizard Agent?

Ask Vizard Agent for a targeted correction and a sync check in the same breath. The second half is the part almost nobody thinks to request, and it is the one that catches the failure you would otherwise have shipped without ever hearing it.

Prompt

Her dialogue drops in the middle section — bring it up to match the rest. Only touch that section, and limit only the peak the boost creates rather than the whole track. Then check the processing has not shifted the audio relative to the picture, and compensate if it has.

Variants worth knowing:

What does Vizard Agent actually do?

Here is the sequence on a real film whose dialogue dipped in one section. The final two steps are the interesting ones: the fix worked, and then the fix had to be checked for a side effect nobody was looking for.

  1. Profiles the audio level across the whole film.
  2. Compares the frequency content of each dialogue section.
  3. Isolates the speaker's level section by section.
  4. Applies the correction to the section that is low.
  5. Diagnoses the unexpected peak the boost creates.
  6. Boosts the low section and limits only the overshoot.
  7. Verifies the dialogue now matches across all sections.
  8. Checks whether the limiter delayed the audio.
  9. Compensates the delay and verifies the rest is untouched.
  10. Muxes and checks the final master specifications.

Step eight is the one worth knowing about. A look-ahead limiter works by delaying the signal so it can see a peak coming, and if that delay is not compensated the whole track arrives late — which shows up as lip sync that is subtly wrong everywhere, from a fix that only touched twelve seconds.

Step six is the difference between a correction and a re-mix. Limiting the whole programme to catch one overshoot flattens the dynamics you spent the edit building; limiting the overshoot alone leaves everything else exactly as approved.

Step three matters because sections differ for different reasons. A speaker who turned away needs a different correction from a section recorded on another day, and the measurement tells you which you have.

What does the result look like?

Dialogue that sits at one level throughout, with the rest of the mix as it was. Vizard Agent reports the level of each section before and after, and confirms the audio still lines up with the picture frame for frame.

When does this not work well?

Level is not the only reason a section can sound wrong. If the quiet passage is also dull or roomy, then raising it simply makes it louder and still wrong, and what it actually needs from Vizard Agent is cleaning rather than boosting.

How do you fix a result that came back wrong?

Say what you hear and where you hear it. Vizard Agent holds the section-by-section measurements it took, so a report like "the middle is now louder than the start" is something it can check against a number rather than a matter of opinion between you.

How does Vizard Agent compare to doing it yourself?

By hand you put a limiter across the master because that is the quick fix, and it solves the peak. The latency it adds is small enough that you do not hear it and large enough that the sync looks slightly wrong, and nobody connects the two.

By hand Vizard Agent
The correction Often across the whole track The section that is low
The peak Caught by a programme limiter Limited on the overshoot only
Latency Not considered Measured and compensated
Verification Listen back Levels compared section by section
Dynamics elsewhere Flattened Untouched

Common questions

Why does a limiter delay the audio? Look-ahead limiters delay the signal to see peaks coming. Vizard Agent checks for it.

How much delay matters? A few milliseconds looks wrong on a close-up, which is why Vizard Agent measures it.

Can it fix just one speaker? Yes. Vizard Agent isolates the speaker's level rather than the section's.

Will the rest of the mix change? No, if you say so. Vizard Agent verifies the untouched sections afterwards.

What if the section is also noisy? Boosting raises the noise too. Vizard Agent will say when cleaning comes first.

Can it hit a loudness target as well? Yes. Tell Vizard Agent the target and the ceiling.

What if the audio was already clipped? Then the detail is gone. Vizard Agent will tell you rather than boosting a flat wave.

Does it re-render the picture? No. Vizard Agent remuxes the corrected audio onto the existing picture.

Can I hear before and after? Yes, and Vizard Agent gives you the numbers as well as the file.

What about the music underneath? It may need ducking instead. Say which you want.

Will it flatten my dynamics? Not if Vizard Agent limits only the overshoot, which is the whole point.

How do I know the sync is still right? Vizard Agent checks it after processing rather than assuming.

Can it do several quiet sections? Yes, each measured and corrected separately.

Is this the same as normalising? No. Normalising moves the whole track; this corrects one section.