iZotope RX 12 De-hum: Remove 50/60 Hz Hum

Use RX 12 De-hum for persistent tonal noise: a 50 or 60 Hz hum, a harmonic ladder or a dense electrical buzz. Choose Static for a simple fixed ladder, learned Dynamic for complex stable interference, or Dynamic with Adaptive enabled for changing hum. Learn from a hum-only passage when possible, then compare on a passage containing the wanted voice or music. Reduce the interference only as far as the source stays natural.

Do not use De-hum as a general “make this cleaner” button. It is a tonal-noise tool. Hiss, wind, room sound and traffic need different processing, even when they occur in the same recording.

The RX 12 De-hum manual, checked September 5, 2026, documents two filter types: Static and Dynamic. Adaptive is an option within them, not a third filter-type tab. Elements includes the De-hum plug-in; Standard and Advanced also provide the standalone Audio Editor. The Editor instructions below therefore require Standard or Advanced.

Confirm that the problem is hum

Hum is persistent tonal noise. In RX's spectrogram it usually appears as a horizontal line at the fundamental frequency with parallel lines above it. Those upper lines are harmonics. A 60 Hz fundamental may show energy at 120, 180 and 240 Hz; a 50 Hz source may produce 100, 150 and 200 Hz. Real interference can deviate from that neat pattern.

Buzz can contain many tonal components extending through the midrange or high frequencies. Rumble is broad low-frequency energy without a tidy ladder; hiss is random broadband noise. Alexey Lukin's 2022 De-hum engineering explanation separates tonal interference from broadband noise. Its listening examples concern older RX versions, while the current manual confirms that RX 12 De-hum handles tonal interference across the spectrum.

Zoom in on a section where the wanted source pauses. If you see stable horizontal rails, De-hum is a strong candidate. If the disturbance is vertical, intermittent or cloud-like, start with the RX spectrogram guide before choosing a module.

50 Hz, 60 Hz or a free frequency?

Power systems make 50 and 60 Hz common starting points, not universal truths. The fundamental you hear may be weak while its second harmonic is louder. A prominent line around 120 Hz can still belong to a 60 Hz hum whose fundamental is partly hidden by a high-pass filter or the recording chain.

In Static mode, adjust Frequency to test 50 or 60 Hz, then confirm the whole ladder. If the spacing points elsewhere, enter that frequency instead. The RX 12 manual and its interface show a Frequency control; older instructions requiring a separate Free selector should not be treated as the current interface. Analog transfers with speed variation can shift hum away from exact mains frequencies.

Do not choose 60 Hz merely because the recording was made in North America, or 50 Hz because it came from Europe. Follow the captured audio. In the vendor's 2023 buzz-removal example, the prominent line is around 120 Hz with several related harmonics. That example is a diagnostic illustration, not proof that every 120 Hz peak is a 60 Hz second harmonic.

Static, Dynamic or Adaptive Dynamic

The current RX 12 product page says Static can address the fundamental plus 16 harmonic bands, while Dynamic can use up to 1,024 dynamically variable notch filters. That large difference is why Dynamic handles dense or inharmonic interference better than a simple harmonic ladder.

Dynamic with Adaptive enabled continuously updates the hum profile. Use it when a learned profile works in one section but misses interference later. The current manual recommends offline plug-in use for Adaptive Dynamic, and the product page warns of high latency. Static also has an Adaptive checkbox for tracking changes, but it remains the simpler harmonic-filter approach; the shared checkbox does not turn Static into Dynamic.

Nine steps to remove hum without hollowing the source

1. Preserve the original. Keep an untouched audio-file copy outside the working edit; a duplicate clip or playlist may reference the same media.
2. Disable Instant Process and inspect. In the Editor, turn Instant Process off before making selections. Inspect the intended channels and identify the tonal pattern.
3. Choose the filter type. Start with Static and Adaptive off for a fixed ladder, or Dynamic for a dense or inharmonic buzz. Enable Dynamic Adaptive only when a fixed profile does not follow the interference.
4. Learn or set the frequency. With Adaptive off, select isolated hum and click Learn in the Editor. In the plug-in, enable Learn and feed hum-only playback, then leave Learn before checking wanted audio. Static Learn sets the base frequency; Dynamic Learn captures the tonal profile. Adaptive Dynamic does not require that fixed profile.
5. Return to wanted audio. Select a representative passage containing voice or music plus hum, including a transition. Start with the full frequency range and intended channels, rather than judging only the learning sample.
6. Limit the necessary processing. In Static, confirm Frequency and enable only justified harmonics. In Dynamic, adjust Bands and the lower and upper Range Selectors to the interference.
7. Adjust and compare. Tune Filter Q and Static harmonic gains or Dynamic Sensitivity. Compare the same original passage at similar perceived loudness; listen for thinning and ringing.
8. Check the removed signal. In the Editor, use the ear-shaped Listen control with Preview, then return to normal output. In a plug-in, use Output Hum Only if available; otherwise compare processed output with host bypass.
9. Apply once and export. Undo all test renders or reopen the untouched working audio, select the intended range and process once. Turn removed-signal monitoring off first. Export to a new filename in the Editor, or render/bounce through the host. Listen through the complete resulting file.

Instant Process applies its selected tool as soon as you make a new selection, which is why it must be disabled before this exercise. For non-destructive listening in the Editor, use Preview and Compare Settings. Render applies processing. Plug-ins do not provide the Editor footer controls apart from Bypass; use host playback and saved alternatives instead.

Set up Static mode conservatively

With Adaptive off, Static places fixed notch filters at the base frequency and its harmonic multiples. Its Learn control can estimate the base frequency from an isolated-hum selection; you can also adjust Frequency manually while previewing. Enable only the harmonics you can justify. RX 12 allows up to 16 above the fundamental, but availability is not an instruction to use all of them.

Link harmonics offers All, Odd/Even and None. All moves the notch gains together; None exposes individual gains, useful when one harmonic is loud and another crosses wanted content. Odd/Even provides two groups. With linking enabled, Slope brings attenuation closer to 0 dB as harmonic order rises. The expanded Harmonic Panel lets you enable or disable individual notches and inspect their gain in dB.

Static works well for a stable mains hum whose lines stay put. Its weakness is equally simple: every notch remains active when wanted content crosses that frequency. A bass note or speech harmonic can disappear along with the hum, leaving a thin or phasey impression.

Static also provides high-pass and low-pass filters ahead of the notch filters. These remove a wider region of the spectrum, so do not treat them as a substitute for Dynamic Range Selectors. Leave them out of the first comparison unless the broad cut is justified. Filter DC offset addresses a waveform offset from zero; it is not another name for removing a 50/60 Hz tone.

Learn a Dynamic profile from the cleanest passage

With Adaptive off, Dynamic learns a tonal profile and changes the notch gains according to the input. In the Editor, select only the interference when possible and click Learn. In the plug-in, enable Learn and play that passage through the active instance; leave Learn before playing the wanted program for evaluation.

If no hum-only passage exists, the current manual allows learning from prominent hum mixed with other audio, but warns that the result may be less useful. Try the least crowded pause, compare several candidate profiles and reject one that removes sustained notes or speech harmonics. Do not switch to Adaptive merely to avoid inspecting the recording; on music it can make the distinction harder.

After learning, preview a section with wanted audio. Increase Bands only when the buzz contains more separate tones than the current setting catches. More filters can follow a complex profile, but every extra intervention creates another chance to touch the program.

The vendor's 2021 Dynamic De-hum walkthrough illustrates that learn-then-listen sequence. Its version-specific performance language is not a guarantee for every RX 12 configuration. If Learn appears to do nothing in a host, check that playback reaches this instance, Learn has finished, processing is not bypassed and your controls address the tonal component. Learning a profile does not itself render a repaired file.

Use Adaptive mode for drift, not convenience

Adaptive Dynamic mode removes the need for a fixed learned profile. It continuously estimates which stable tones belong to hum and updates as their frequency or amplitude changes. This is useful for spliced recordings, motors, generators and transfers whose speed is not perfectly stable.

Adaptive Dynamic needs lookahead, and the current manual explicitly recommends offline use in the plug-in. Use the Editor or the host's offline processing route, then verify the returned timing. Do not use the roughly two-second delay described in the 2022 engineering article as a measured RX 12 latency for your system.

Be especially careful on music. iZotope's engineering explanation says Adaptive was formulated primarily for speech and may classify sustained musical notes as hum. Compare it with a learned Dynamic profile and listen to held bass, organ, strings and guitar notes from attack through decay.

Filter Q: narrower is not always safer

Filter Q controls notch bandwidth. A higher Q makes a narrower notch, which seems ideal because it removes less neighboring spectrum. The tradeoff is time-domain ringing: very narrow filters have a longer impulse response and can leave a resonant tail around transients.

Begin narrow enough to isolate the visible line, then listen around plosives, drum hits and edit boundaries. If a metallic or resonant after-sound appears, reduce the Q or try Dynamic mode so the notch can release when wanted energy dominates.

In Static mode, Linear-phase filters can add latency and pre-ringing. With the option off, minimum-phase filtering places ringing after the transient, where the source may mask it better. The current manual describes this trade-off and notes that disabling Linear Phase reduces real-time plug-in latency. It does not promise zero end-to-end delay or an artifact-free result.

Sensitivity, Bands and Range in Dynamic mode

Sensitivity changes how readily RX treats detected tonal energy as hum. Raise it until the objectionable line is controlled, then stop. If the source loses body or Hum Only contains recognizable program, lower it.

Bands determines how many dynamic notch filters are available. A simple hum needs fewer than a dense electrical buzz. Increasing the count can capture more interference at the expense of transparency.

Range Selectors define the lower and upper processing limits. Use them to protect a region that contains no hum or to stop a high-frequency profile from touching diction and cymbals. The range should follow visible evidence, not a generic speech or music preset.

Change one control at a time and keep the same loop. Otherwise a disappearing harmonic can make an unrelated adjustment seem successful.

Hum Only is the fastest artifact check

The RX 12 manual calls removed-signal monitoring Output Hum Only; its current Editor images show the ear-shaped Listen control beside Compare in both filter types. Enable it and use Preview to hear what processing removes. A useful result is dominated by the interference. If clear words, notes or attacks emerge, compare weaker settings and judge normal output again; faint leakage alone is not a reason to chase a perfectly silent difference signal.

Return Listen or Output Hum Only to normal output before rendering or bouncing. The common-controls manual says the Editor footer is absent from plug-ins apart from Bypass, so do not search a host instance for the same Preview/Compare/Listen row. If your plug-in exposes Output Hum Only, use it; otherwise rely on processed-versus-bypassed playback. Match perceived loudness so a level change does not disguise lost voice body or bass.

Check on headphones for ringing and on ordinary speakers for missing weight. A repair that sounds spectacular on a subwoofer but makes speech thin on a laptop has not passed.

De-hum does not remove every kind of noise

After a successful pass, broadband hiss may remain. That is expected. De-hum targets tonal components; it does not promise a silent background. Use Voice De-noise or Spectral De-noise for the random component, and keep both processes conservative.

The Voice De-noise versus Spectral De-noise guide explains that choice. For changing speech backgrounds and room reflections, compare Dialogue Isolate. For a single horizontal intrusion, a local frequency selection and Spectral Repair may avoid processing the full duration. Spectral De-noise, Dialogue Isolate and Spectral Repair require Standard or Advanced; Elements users have the Voice De-noise plug-in.

A harmonically rich buzz can sometimes behave like a train of periodic clicks. iZotope's engineering guide suggests De-click as an alternative in that special case. Diagnose the shape before stacking every denoiser.

Common De-hum mistakes

Removing the loudest visible line as the fundamental. It may be the second harmonic. Check spacing below and above it.

Using all 16 Static harmonics. Disable notches that do not correspond to real interference.

Learning from dense music or speech. Find a cleaner pause or switch to Adaptive only when drift justifies it.

Using Adaptive on sustained music without comparison. Held notes can resemble unwanted steady tones.

Maximizing Q, Bands or Sensitivity. More precision on a display can mean more ringing or program loss in playback.

Chasing absolute silence. Stop when the hum no longer distracts in context. Stable low-level residue may be preferable to a hollow voice or damaged bass.

A final De-hum quality gate

• The fundamental is confirmed from the harmonic spacing, not geography alone.
• The selected mode matches simple, complex or drifting interference.
• Only necessary Static harmonics or Dynamic bands are active.
• Hum Only is dominated by interference rather than wanted program.
• Plosives, drum attacks and edits do not produce obvious ringing.
• Bass notes and vocal body survive matched-level bypass comparison.
• The setting works on several sections, not only the learning sample.
• Remaining hiss or room tone is handled separately only if it still matters.

De-hum belongs early in a repair chain because a strong tonal ladder can confuse broader denoisers. But diagnose clicks and clipping before learning a noise profile when those events dominate the sample. The Repair Assistant guide helps when the defect is still unclear, and the Repair Audio hub routes isolated problems to their dedicated modules.

For the final deliverable, File > Export writes an audio file such as WAV or AIFF. Ordinary Save overwrites an open WAV or AIFF; Save RX Document preserves editable RX history in an .rxdoc that opens in RX. Use a new audio filename for delivery, reopen it and check duration, channel layout, sync, peaks and transitions across the entire program. A quiet learning sample is not a finished repair.

Frequently asked questions

Should I choose 50 Hz or 60 Hz?

Use the value that matches the captured fundamental and harmonic spacing. In Static mode, adjust Frequency or use Learn to estimate it. Enter another frequency if the evidence points elsewhere; a separate Free selector is not required by the current RX 12 interface.

What is the difference between Static and Dynamic De-hum?

With Adaptive off, Static uses fixed notches at a fundamental and its harmonics. Dynamic learns a more complex tonal profile and changes notch gains with the input. Adaptive is a checkbox within the filter types, not a third tab.

When should I use Adaptive Dynamic mode?

Use Dynamic with Adaptive enabled when changing interference defeats a fixed learned profile. The current manual recommends offline plug-in use. Compare carefully on sustained music, which can be mistaken for hum.

How many harmonic bands should I enable?

Enable only the Static harmonics that correspond to visible and audible interference. RX 12 provides 16 above the fundamental, but using every one can remove wanted content.

What does Filter Q do?

It sets notch bandwidth. Higher Q is narrower, but extremely narrow filters can ring around transients. Adjust while previewing speech, music and edit boundaries.

Why does De-hum make the voice or bass sound thin?

Notches may overlap wanted harmonics, or Dynamic Sensitivity and Bands may remove too much. Disable unnecessary Static harmonics, compare Filter Q settings and judge processed output against bypass. In the Editor, Listen with Preview helps reveal what is being removed.

Can De-hum remove hiss too?

Not by design. De-hum targets tonal interference. Use Voice De-noise or Spectral De-noise for broadband hiss after the tonal hum is controlled.

Is De-hum included in RX 12 Elements?

Yes. De-hum is included in Elements, Standard and Advanced. Elements supplies it as a plug-in; Standard and Advanced also include the RX Audio Editor workflow.