Skip to guide
← Tool guides

Audio, explained · 13 / 36

Denoise

Learn a background and reduce similar spectral noise.

Visual walkthrough3 configuration controlsSynced audio walkthrough
See how it works ↓

During learning, audio passes unchanged. Learning needs a running, enabled, non-bypassed tool. The algorithm uses 1024-sample analysis frames and declares 1024 samples of latency.

01 / Understand

The idea, before the settings.

Learn a background noise profile while the wanted voice is absent, then reduce spectral content matching that profile. Reduction sets strength and Noise left sets a minimum gain floor.

Steady noisy micDenoiseCompressor → Output

Use it in AudioRouter

  1. Play the route and keep the wanted source quiet.
  2. Click Learn noise while only the steady room/fan background is present, then Stop learning and keep profile.
  3. Resume speech and adjust Reduction. Raise Noise left if speech becomes watery.
  4. Relearn after moving the mic or changing the background.

What success looks like

The learned background drops, including under phrases, without excessive damage to wanted detail.

If it does not work

If the profile contains speech, learn it again in silence. Increasing Reduction cannot repair a contaminated profile.

Select the node to open Properties. Click its name to rename it. While playing, supported parameter-only edits save after a short pause. For structural edits, use Save when the canvas marks a change. Enabled and Bypass are distinct live controls; see the in-app help for their behavior.

02 / Explore

Make the behavior visible.

The lesson starts with a prepared synthetic hiss profile. It attenuates a separately known noise layer, not a real spectral-subtraction implementation; use it to compare reduction and residual noise.

The example repeats until you press Stop, and control changes are heard immediately. The picture shows the tool's shape rather than time; follow the timing in Follow the sound below. This is not a DSP measurement or control for your running AudioRouter session.

03 / Configure

Every control, with a reason.

01

Reduction

reductionPercent

Strength of learned-profile spectral subtraction.

Range / choices0 % → 100 %Default70 %

↓ Lower / alternativeRetains more original sound.

↑ Higher / alternativeRemoves more of the learned spectral background.

Try thisStart at 70% after learning room noise.

Keep in mind High values can create watery or musical artifacts.

02

Noise left in

floorPercent

Minimum gain permitted in each spectral bin.

Range / choices0 % → 100 %Default10 %

↓ Lower / alternativePermits deeper spectral attenuation.

↑ Higher / alternativeRetains more background and may sound more natural.

Try thisTry 20% if a 10% floor sounds too hollow.

Keep in mind This is a gain floor, not a guarantee of a particular final noise percentage.

03

Learn noise

learning

Runtime learning controlled by Learn and Keep profile actions.

Range / choicesOn / offDefaultfalse

↓ Lower / alternativeStop and keep the measured profile before evaluating speech.

↑ Higher / alternativeStart learning only while the unwanted background is present.

Try thisStay quiet while the fan runs, then keep its profile.

Keep in mind Requires a running enabled non-bypassed tool; the profile is stored by the editor rather than typed as raw hex.

Beyond the controls

Actions & transport

Learn noise

Start while the route is running, the tool is enabled and not bypassed, and only unwanted background is present.

Stop learning and keep profile

Stores the live 64-band profile and leaves learning mode. The editor writes the bounded profile; users do not type its 128-character hexadecimal representation.

04 / Put it to work

Start with a useful example.

Example 1

Learn room noise

  1. Start the route and stay quiet.
  2. Click Learn noise while the fan or room background is present.
  3. Click Stop learning and keep profile, then speak and adjust Reduction.
Example 2

Reduce watery artifacts

  1. If speech becomes watery, lower Reduction.
  2. Raise Noise left in slightly, for example from 10% to 20%.
  3. Relearn after moving the microphone or changing the room background.

These are starting points. Compare at a comfortable monitoring level and adjust for your own source.

05 / Get unstuck

Small details that matter.

  • Speaking during learning can teach the tool to remove parts of your voice.
  • Noise left in at 0% permits deeper suppression; it does not guarantee every unwanted sound is removed.