What is the difference between Balanced, Studio and Ultra, then?

Balanced: good quality with faster processing, at 0.5x credits. Studio: high quality for professional use, at 1x credits. Ultra: maximum quality with longer processing time, at 2x credits. Those are the tool's own descriptions, and they are about processing and time rather than about how much gets taken out.

Which one should I choose?

Start at Studio, which is where the menu already is. If the result is clean enough, stop — a cheaper or dearer run is not going to make a good result better in a way you will notice. If it is not clean enough, or if it came back dull, run a short excerpt at another setting and compare rather than committing a long recording to a guess.

How do I know it removed too much?

By listening to the voice rather than to the background. There are five tells: consonants soften, the voice goes distant, quiet moments swim, breaths disappear and the background pumps. Any one of them means the cleanup went past the point where it was helping.

Can I run it twice to get it cleaner?

You can — the output of one job can be uploaded as the input to another — but it is usually the wrong move. Each pass takes something away, so a second pass on an already-clean file mostly removes voice. It also costs the same as the first. If one pass did not solve it, the problem is usually the recording rather than the number of passes, which is the next lesson.

Does a re-run cost again?

Yes. A re-run is a new job at the same rate — 3,300 credits for a five-minute clip at Studio, 79,200 for a two-hour one. That asymmetry is why the excerpt-first habit is worth having on anything long.

Noise Reduction

How much to remove

Q: Where is the strength setting?

There isn't one. Noise Reduction has a single Quality menu — Balanced, Studio and Ultra — and it is not a strength dial. Nothing in the tool asks how much noise to take out — the three options change how much processing the recording gets and what it costs, not how aggressive the cleanup is. If you have read elsewhere that you can dial the cleanup up and down, that is not what the tool in front of you does.

Cleanup takes the noise and some of the voice with it. What the Quality menu really controls, how to hear an over-processed recording, and why the lightest result that solves your problem is the one to keep.

Lesson 2 · core · 7 min read

Open Noise Reduction

Noise reduction works by taking things away, and it cannot see the difference between a fan and the top of somebody's voice as clearly as you can. So every cleanup is a trade: the further it goes, the less noise you have and the less voice you have. A recording can be perfectly quiet and still be worse than the noisy one you started with.

Careful

It is not a strength dial. Nothing in the tool asks how much noise to take out — the three options change how much processing the recording gets and what it costs, not how aggressive the cleanup is. If you came here looking for a low, medium and high setting, it does not exist — say so to yourself now rather than hunting the interface for it.

Option: What the tool says, and what it costs
Balanced: Good quality with faster processing — 0.5x credits.
Studio: High quality for professional use — 1x credits.
Ultra: Maximum quality with longer processing time — 2x credits.

The Quality menu, in the tool's own words.

Studio is where the menu starts. Balanced is the cheap end at 0.5x credits and Ultra the dear end at 2x credits, so on a one-hour recording the spread runs from 19,800 credits to 79,200. Nothing in the tool predicts which of them will sound best on your recording, and this guide will not pretend otherwise — it is a thing you find out by hearing it.

What too much sounds like

The noise leaving is easy to hear and it is not the thing to listen for. What you are listening for is what left with it, and it shows up in five recognisable ways.

Consonants soften. The s, t and f sounds lose their edge, and speech starts to feel like it is being described rather than heard.
The voice goes distant. It sounds as if a blanket went over the microphone. The noise left and it took the air with it.
Quiet moments swim. Words at the end of a sentence, where a voice trails off, warble or drop out entirely.
Breaths disappear. A voice with no breaths in it reads as synthetic even when every word is intact.
The background pumps. Instead of a steady hiss you get silence that opens up whenever someone speaks and closes again when they stop.

Tip

Listen to a sentence, not a word. Over-processing shows up most at the ends of phrases, where a voice trails off quietly and the cleanup can no longer tell it from the background. A word in the middle of a loud sentence will sound fine in almost any version.

Write this

Not this

What you play back

The same twenty seconds of both, one after the other

The cleaned file on its own

What you listen for

Whether the voice still has its edges and its air

Whether the noise is gone

Which version you keep

The lightest one that solves the actual problem

The quietest one

A two-hour recording

Run two minutes from the middle first, then commit

Run the whole thing and hope

Still noisy after one pass

Ask whether the recording itself is the problem

Run the output through again

What you play back

Write this:: The same twenty seconds of both, one after the other
Not this:: The cleaned file on its own

What you listen for

Write this:: Whether the voice still has its edges and its air
Not this:: Whether the noise is gone

Which version you keep

Write this:: The lightest one that solves the actual problem
Not this:: The quietest one

A two-hour recording

Write this:: Run two minutes from the middle first, then commit
Not this:: Run the whole thing and hope

Still noisy after one pass

Write this:: Ask whether the recording itself is the problem
Not this:: Run the output through again

Test on an excerpt, not on the whole thing

A re-run is a new job at the full rate, so the cost of being wrong scales with the length of the recording and nothing else. Two minutes from the middle of a long file costs about 1,320 credits at Studio and answers the only question that matters: does a cleanup make this particular recording better?

Note

Take the excerpt from the middle. The first minute of any recording is the least representative part of it — people settle, rooms settle, and the noise you are actually fighting usually turns up once everyone has stopped being careful.

The lightest thing that solves it

The goal is not the cleanest possible file. It is a recording somebody can listen to without noticing anything — and a listener notices a hollow, underwater voice far faster than they notice a quiet hiss. Broadcast and podcast audio is full of low-level room tone that nobody hears, because it sits under speech that still sounds like speech.

Careful

If none of the settings give you something better than the original, keep the original. That is a real outcome, not a failure to try hard enough, and it is the honest answer more often than a tool's own guide will usually admit.

When a cleanup will not get there no matter what you choose, the reason is almost always the recording rather than the setting — and there is a short list of recordings this tool cannot rescue at all. That is the next lesson, and it is the one worth reading before you spend anything.

Questions

What people ask about this

Free to start

Now go make one

Reading about a style description only gets you so far. The studio is free to use — write one sentence and hear what comes back.

Create a free account

1,000 credits every month. No card required.

All Noise Reduction lessons

What Quality actually controls

Careful

Option: What the tool says, and what it costs
Balanced: Good quality with faster processing — 0.5x credits.
Studio: High quality for professional use — 1x credits.
Ultra: Maximum quality with longer processing time — 2x credits.

The Quality menu, in the tool's own words.

What too much sounds like

The noise leaving is easy to hear and it is not the thing to listen for. What you are listening for is what left with it, and it shows up in five recognisable ways.

Consonants soften. The s, t and f sounds lose their edge, and speech starts to feel like it is being described rather than heard.

The voice goes distant. It sounds as if a blanket went over the microphone. The noise left and it took the air with it.

Quiet moments swim. Words at the end of a sentence, where a voice trails off, warble or drop out entirely.

Breaths disappear. A voice with no breaths in it reads as synthetic even when every word is intact.

The background pumps. Instead of a steady hiss you get silence that opens up whenever someone speaks and closes again when they stop.

Tip

Write this

Not this

What you play back

The same twenty seconds of both, one after the other

The cleaned file on its own

What you listen for

Whether the voice still has its edges and its air

Whether the noise is gone

Which version you keep

The lightest one that solves the actual problem

The quietest one

A two-hour recording

Run two minutes from the middle first, then commit

Run the whole thing and hope

Still noisy after one pass

Ask whether the recording itself is the problem

Run the output through again

What you play back

Write this:: The same twenty seconds of both, one after the other
Not this:: The cleaned file on its own

What you listen for

Write this:: Whether the voice still has its edges and its air
Not this:: Whether the noise is gone

Which version you keep

Write this:: The lightest one that solves the actual problem
Not this:: The quietest one

A two-hour recording

Write this:: Run two minutes from the middle first, then commit
Not this:: Run the whole thing and hope

Still noisy after one pass

Write this:: Ask whether the recording itself is the problem
Not this:: Run the output through again

Test on an excerpt, not on the whole thing

Note

The lightest thing that solves it

Careful