How to make an amateur performance recording sound better
Upload the recording and tell Vizard Agent the singing and the music both need help. Vizard Agent measures the loudness of the speech and the song sections separately, sets the limiter from the recording's own peaks, brings the voices forward without thinning them, and tidies the cuts to real shot changes.
What is the short version?
A recording of a live performance made on a phone has two different audio problems in one file: the spoken sections and the sung ones need opposite treatment, and processing the whole thing uniformly ruins one to fix the other.
- Go to Vizard Agent and upload the recording.
- Say the singers' voices and the music both need improving.
- Say whether the cuts should be tidied as well.
What do you need before you start?
The original recording, not a re-exported copy. Vizard Agent measures the audio itself and decides the treatment from those measurements, so no technical description is needed from you. Knowing which parts matter most does help it set the priorities.
- The recording. As captured, not re-exported.
- The priority. Voices, music, or both equally.
- The sections. Speech, singing, or mixed.
- The cuts. Whether to tidy them too.
- The use. Social, a keepsake, an archive.
What do you type into Vizard Agent?
Name the parts separately. Vizard Agent measures speech and singing on their own terms, and telling it that the recording contains both is what stops a single noise-reduction pass being applied across material that needs two different settings.
Prompt
Variants worth knowing:
- Voices forward. The singing lifted above the backing.
- Noise controlled, not removed. Enough to help without thinning.
- Cuts on real changes. Rather than on a fixed rhythm.
What does Vizard Agent actually do?
Here is the order Vizard Agent worked in on a real amateur performance recording, from the first look at the file to the final level decisions. The stretch from the thirteenth step is the careful part: it measured the spoken and the sung sections separately before touching either one.
- Looks at the video's content, duration, picture quality and audio character.
- Prepares previews of several sections to judge the framing.
- Checks the rhythm of the shot changes and the clarity of the sound.
- Analyses the voices, the music and how the audio relates to the picture.
- Measures the loudness and peaks of the original so levels can be set properly.
- Confirms where the black screen starts and checks the key moments.
- Maps the shot changes so the visual cuts can be tidied.
- Refines the cut points against real frame changes rather than a fixed interval.
- Measures the spoken section's loudness so noise reduction will not thin it, measures the song section so the singers can come forward, and reads the digital peaks to set the limiter.
Step 9 is the whole method. Noise reduction strong enough to clean a spoken passage will hollow out a sung one, and a limiter set from a guess will either do nothing or squash the loudest notes — both are avoided by measuring rather than estimating.
What does the result look like?
From the run this page is written from, probed on the delivered file: 576x1024, H.264, 30fps, 229.37 seconds, AAC audio. Just under four minutes of the performance with the singers audible above the backing, the noise controlled without the voices going thin, and the cuts landing on actual shot changes.
The picture is the same footage. What changed is that it is now listenable end to end, which for a recording of a real occasion is usually the difference between something people watch and something nobody opens twice.
When does this not work well?
Audio repair can only work with what was actually captured, and a phone microphone at the back of a hall has limits that no amount of processing removes. These are the ones worth knowing before you expect too much of the result.
- Clipping is permanent. Distortion recorded into the file cannot be undone.
- One microphone hears one place. Distance and room are baked in.
- Heavy noise reduction thins voices. There is a real ceiling on how far it goes.
- Bad picture stays bad. Audio repair does not improve the footage.
- Live is live. A wrong note stays a wrong note.
How do you fix a result that came back wrong?
Say which section sounds wrong to you and how. Vizard Agent keeps all the loudness measurements, the peak readings and the shot map, so the speech or the singing can be re-treated separately without redoing the whole file.
- "The voices sound thin." Noise reduction eased on that section.
- "The music is louder than the singer." Re-balanced from the measured levels.
- "That cut is jarring." Moved to the nearest real shot change.
How does Vizard Agent compare to doing it yourself?
By hand this means an audio editor, a noise-reduction plugin and a lot of guessing, applied to a whole file at once because splitting it by section is tedious. The result is usually a clean spoken part and hollow singing.
| By hand | Vizard Agent | |
|---|---|---|
| Treatment | One pass over everything | Speech and singing measured separately |
| Levels | Set by ear | Loudness and peaks measured |
| The limiter | A default setting | Set from the recording's own peaks |
| Cuts | Trim on a rhythm | Placed on real shot changes |
Common questions
Can Vizard Agent fix a phone recording of a concert? It can improve it considerably. It cannot undo distortion or move the microphone.
Will the singers sound better? They will sit further forward. Vizard Agent measures the song sections to lift them.
Why treat speech and singing differently? Because the settings that clean one hollow out the other, and one pass over both always sacrifices something.
Can it tidy the video cuts too? Yes. Vizard Agent maps the shot changes and moves cuts onto real ones.
What about clipping? Unfixable, unfortunately. Vizard Agent works around it and tells you where it occurs.
How does it decide the limiter setting? From the recording's measured digital peaks, rather than from a default that may be far off.
Does the picture change? Only the cuts, and only if you ask. Vizard Agent does not regrade anything unless told to.
Can it separate the vocals from the backing? To a degree. Say so and Vizard Agent attempts a separation before balancing.
How long a recording can it handle? Any length. Longer recordings simply mean more sections to measure.
Is it worth doing on an old family recording? Often more than on a recent one. Vizard Agent cannot improve the picture much, but making decades-old audio listenable is usually what decides whether anyone in the family watches it again.
Should I ask for the audio on its own as well? Often worth it. Vizard Agent can deliver the treated track separately, which is useful if the recording is going to be re-edited later by somebody working from the original picture.
What is the single biggest improvement? Getting the voices above the backing. Vizard Agent measures the sung sections specifically for this, and a performance where the singer is finally audible reads as a different recording entirely.
Does Vizard Agent check the result? Yes. It reviews the levels and the assembled cut before delivery.