How to Remove Reverb in iZotope RX 12

To reduce room reverb in iZotope RX 12, use Learn on a passage containing direct sound and clear decays, with a little room tone at the beginning. Refine Reverb Profile and Tail Length, then increase Reduction cautiously. If the voice becomes hollow or word endings lose detail, back off. Check Output Reverb Only alongside the processed voice: the aim is less distracting room without damaging the performance.

Use the general De-reverb module when the source is music, mixed material or a room that benefits from a learned profile. For speech, compare it with the current Dialogue Isolate workflow, which separates dialogue, reverb and noise. Neither tool can recreate a close microphone from a distant recording, but both can make a usable take more focused.

De-reverb is available as a plugin in Elements, Standard and Advanced. The standalone RX Audio Editor used below requires Standard or Advanced; Elements does not include it. This distinction is confirmed in the current RX 12 edition tables. In a DAW plugin, audition with host playback and bypass; the Editor's Compare and Render buttons are a separate workflow.

If the real-time plugin struggles during playback, test an offline render or the Editor route available in your edition. The manual notes that this processing can be demanding. For installation, use the current RX 12 release requirements: the module page retains an old RTAS format reference that the current release list does not support. Neither a universal latency figure nor better sound merely from moving to the Editor should be assumed.

First decide whether the problem is reverb

Reverb is the dense trail of reflections that follows direct sound in a room. A voice recorded far from the microphone may smear from one syllable into the next. In the spectrogram, sharp events become blurred forward in time instead of ending cleanly.

Do not confuse that with a distinct repeated echo, steady hiss or electrical hum. An isolated reflection may be worth trying as a local repair, but an echo overlapping the next word is much harder to separate. Broadband noise needs a denoiser; a fixed harmonic ladder may call for De-hum. The RX spectrogram guide helps interpret those shapes alongside listening.

Reverb is not automatically bad. It supplies scale, distance and continuity. The problem is excess room that masks diction, conflicts with the production space or makes several recordings sound as if they happened in different buildings.

What RX De-reverb actually models

De-reverb estimates the relationship between reverberant and direct sound by frequency, plus the decay rate of the room. Learn uses that analysis to suggest Reverb Profile and Tail Length settings. The RX 12 De-reverb manual describes this wet/dry model; Learn does not decide how much damage you are willing to tolerate.

Those values are a starting point, not a verdict. The algorithm cannot know whether a quiet tail is unwanted room, intentional production reverb or part of the instrument. Your listening test decides what survives.

The useful goal is usually reduced distance rather than absolute dryness. A believable voice with a faint stable room is easier to mix than a brittle voice surrounded by moving digital residue.

Nine steps for natural reverb reduction

1. Preserve the source. Make an independent copy of the audio file and keep the original untouched; a duplicated DAW region may still reference the same file.
2. Prepare a diagnostic selection. In the RX Audio Editor, turn Instant Process off before selecting audio. Include direct sound, quiet words and complete room tails from one recording condition.
3. Choose the tool. Open De-reverb for a learned general room profile. For spoken dialogue, compare Dialogue Isolate separately on the same untouched passage.
4. Learn representative audio. Select about five seconds that starts with room tone and includes the source and its decays, then click Learn. Try a longer consistent passage if needed.
5. Refine profile and tail. Preview the suggestion, adjust only the frequency bands and Tail Length that need correction, and compare quieter as well as louder phrases.
6. Set reduction and check the difference. Increase Reduction cautiously. Use Output Reverb Only to investigate lost attacks, then turn it off and confirm that the processed voice retains detail.
7. Compare and render one approved pass. Use Compare for alternatives on the same unprocessed selection. Undo all test renders or reopen the untouched copy before applying the chosen settings once, with Output Reverb Only off.
8. Try a second pass only if needed. Retain the first-pass result, Learn its remaining room again and test a shorter-tail treatment for distracting early reflections. Reject it if the first pass sounds more natural.
9. Export and approve in context. Export a new audio filename, reopen it and check the full duration, quiet endings, headroom and edit continuity at matched listening level in the destination mix.

The Editor Compare window lets you name candidate settings and audition their processed results. Compare alternatives against the same input: a second pass is processing an already changed file, so keep a separate first-pass reference. Preview does not commit the repair; Render does. Instant Process must stay off while you diagnose, because making a selection can otherwise apply an unrelated repair immediately.

Choose a useful Learn selection

Start with about five seconds from one acoustic condition. The selection should begin with a little room tone, then include direct signal and audible tails. This differs from learning a denoiser from noise alone. A clap, cough, drum hit or phrase with spaces can expose decay more clearly than continuous sustained audio. If the estimate is poor, the manual suggests trying another passage or extending analysis, sometimes up to ten seconds; these are starting suggestions, not fixed requirements.

Avoid learning across a hard room change. If half an interview was recorded in a kitchen and half in a bedroom, build separate processing states. One averaged profile may fit neither environment.

Run Learn and preview the suggestion. Audrey Martinovich's 2024 iZotope De-reverb walkthrough follows the same basic progression: analyse the room, refine its tonal and temporal shape, then choose the strength. Its example is useful technique, not a newly measured RX 12 performance result.

Read the Reverb Profile by ear and frequency

The Reverb Profile divides the correction across frequency bands. A hard small room may ring brightly; a larger resonant space may build mud in low and low-mid bands. Increase a band only when its decay is disproportionately strong.

Watch for mistaken identity. Low vocal fundamentals, bass sustain, cymbal wash and sibilance can overlap the room's frequency range. Audition Output Reverb Only after a profile change, then return to normal output. If a stronger low-band setting makes the processed singer lose body, or a high-band change weakens consonants, reduce that correction. Hearing pitch in reverberation alone is not proof of damage.

The profile should describe the room, not redraw the source's tone. Correct final tonal balance later with EQ if necessary; do not force De-reverb to act like a broad equalizer.

Set Tail Length from the audible decay

Tail Length sets the decay behaviour of the processing. The manual describes it as an approximation of RT60: the time for reverberant sound to fall by 60 dB. Learn estimates a value, but exposed phrase endings help you refine it. Treat this as a processing control, not a calibrated room-acoustics measurement from an arbitrary recording.

If long tails remain after otherwise clean processing, try a longer value. If word endings become dull, the noise floor churns or the entire voice seems suppressed after each syllable, shorten it. Very short settings emphasize early reflections, the rapid first bounces that often occupy roughly the first 5–100 milliseconds.

Do not measure only the most dramatic empty pause. Use several phrases at different levels. Quiet speech can reveal overprocessing that a loud sentence hides.

Reduction: find the damage threshold, then step back

Reduction is the main wet-component control. Larger values remove more of what the model identifies as reverb. Negative values can enhance rather than reduce the reverberant component, so confirm the direction before rendering.

A practical method is to increase Reduction until you first notice hollow vowel tone, fluttering room texture, shortened consonants or unstable ambience between words, then lower it again. The original may still sound roomier in solo. Check the processed voice through the compression and mix you intend to use; a setting that sounds acceptable alone may fail there.

Do not approve the louder option by accident. Match listening levels between processed and bypassed versions. Enhance Dry Signal increases the direct component; check output peaks when using it, because added gain is not evidence that separation improved.

Artifact Smoothing and Enhance Dry Signal

Artifact Smoothing changes the frequency precision of the processing. Its default is high because reverberation is generally smooth across frequency. Lower values may help with narrow resonances, but can make strong reduction sound more artificial. Start from the learned profile with restrained Reduction; high smoothing does not make every aggressive setting safe.

Change one control at a time. If a resonant room tone remains, lower smoothing slightly and compare. If the output becomes grainy or musical, restore smoothing or reduce the main amount.

Enhance Dry Signal raises the direct component and can help speech or transient material feel more forward. Use it only after the reverb model behaves correctly. It cannot repair an inaccurate profile; it can merely make an overprocessed direct signal louder.

Use Output Reverb Only to investigate lost detail

Output Reverb Only auditions the component RX intends to remove. It may not resemble a natural standalone reverb because it represents the processing difference, but it should still tell you whether the model is stealing content.

Reflections contain the same words, pitches and rhythms as the direct performance, so the removed signal need not be unintelligible. Be more concerned when sharp consonant fronts or complete attacks dominate it and those details have become weaker in the processed output. Compare the same quiet phrase with the original before blaming a particular control.

This monitoring step is more reliable than staring at a cleaner spectrogram. A visually sharp syllable is not an improvement if it has lost the information that makes it understandable.

Turn Output Reverb Only off before normal listening and the approved render. Leaving it on changes the output to the residual reverberant component rather than the repaired recording. Keep listening volume comfortable when switching between the quiet residual and the full signal.

One hard pass or two light passes?

The manual describes an optional long-tail pass followed by a newly learned, shorter-tail pass for difficult early reflections. Its example uses Tail Length 0.5 seconds and Artifact Smoothing around 3.0 for that second stage. Treat those as an experiment on a retained first-pass copy, not universal vocal settings. A quieter, simpler single pass may already do the job.

Before testing the second stage, keep a separate first-pass file or an RX Document with its history. Learn the first-pass result again, then judge the extra treatment against that first-pass reference as well as the original. Do not render the first stage twice by accident. Each stage needs to solve a remaining audible problem without adding a worse one.

Stop when the recording sits in context. Absolute dryness is rarely worth a metallic or phasey voice. Severe distance and reflections that overlap every syllable may require re-recording, ADR or a creative production decision rather than more reduction.

General De-reverb versus Dialogue Isolate

The current Dialogue Isolate manual separates Voice, Reverb and Noise with independent gain controls in Standard and Advanced. Advanced adds Best / Offline quality and multiband controls. Increasing Sensitivity sends more signal into noise and reverb, potentially losing voice detail. That Sensitivity control belongs to Dialogue Isolate; general De-reverb uses the learned profile, Reduction and Tail Length instead.

Use the Dialogue Isolate guide when the job is primarily voice-versus-background. Use this page when a learned room profile and decay controls are the center of the repair.

Older tutorials also mention a separate Dialogue De-reverb module. In his RX 12 vocal-cleanup article, Nick Messitte says it remains accessible through key commands in his setup. That does not make its legacy sensitivity, ambience or stereo controls part of the general De-reverb panel. The current documented speech route here is Dialogue Isolate; check your installed build before following a legacy-module shortcut.

De-reverb before or after noise reduction?

There is no universal order. The De-reverb manual permits either order with Spectral De-noise. Compare the two routes on independent copies when both problems matter: one with modest denoising before room reduction, another with room reduction first. Use the same phrase, similar listening level and similarly restrained amounts, then choose by the remaining noise, room continuity and vocal damage.

Obvious clicks, clipping and plosive bursts may merit targeted repairs before broad processing. That is a diagnostic starting point, not a reason to run every module over every file. The background-noise workflow separates fault types, while the vocal-cleanup chain puts room treatment in the context of a complete performance.

When both noise and reverb are heavy, alternate small corrections and monitor what each stage removes. Stacking maximum De-reverb and maximum denoising is a common route to watery speech.

A discussion about combining RX separation and denoising asks these same order-of-processing questions. The visible replies describe individual preferences and overprocessing problems, rather than a controlled comparison. Use that as a reason to test your material, not proof that more modules will produce a better voice.

Why De-reverb sounds hollow, metallic or gated

Hollow vowels: Reduction or a profile band may be treating useful sustain as room. Back off, check the removed signal, then confirm that the processed voice regains its body.

Metallic or watery texture: Strong reduction, low smoothing or an unrepresentative Learn selection can contribute to artifacts. Relearn a consistent section and lower the amount, comparing one change at a time.

Gated endings: Recheck Reduction, Tail Length and the learned profile on quiet syllables. Reduce the amount or shorten an overlong tail setting; do not look for a Sensitivity slider in general De-reverb.

Room pumps between phrases: Preserve more consistent room and reduce the processing. In Dialogue Isolate, balance its Reverb and Noise gains rather than forcing both to silence; review the ambience across edit boundaries.

Voice is clearer but unnaturally close: The tool succeeded technically and failed aesthetically. Restore some reverb or blend an appropriate production space later.

Final quality gate

• Processed and original versions are compared at matched loudness.
• Sharp direct attacks do not disappear from the repaired audio; Output Reverb Only has been checked and switched off.
• Long tails are reduced enough for the production, not erased for appearance.
• Early reflections no longer mask clarity without shortening consonants.
• Room tone remains stable through pauses and edit boundaries.
• No hollow, metallic, watery or gated artifact appears on quiet phrases.
• Noise reduction solves noise separately rather than forcing De-reverb harder.
• The result works in the complete mix on speakers and headphones.

For delivery, use File > Export with a new filename such as interview-dereverb-v1.wav. Match the destination's required format and check the result after reopening it. Export Selection only delivers the selected excerpt. Ordinary Save can overwrite an open WAV or AIFF, so keep the untouched original separate.

An RX Document preserves original audio and edit history for revisiting the repair, but it is not the audio file you send to a DAW. After importing the exported audio, check duration, alignment, channel layout, peaks and the joins with surrounding room tone. For other speech restoration routes, continue through the Clean Dialogue hub.

Frequently asked questions

Can iZotope RX completely remove reverb?

Sometimes it can reduce reverb dramatically, but complete transparent removal is not guaranteed. Dense reflections that overlap every syllable may leave artifacts before the room disappears.

How long should the De-reverb Learn selection be?

Use several representative seconds containing room tone, direct signal and clear tails. Longer is useful when the decay is complex, but do not cross into a different acoustic condition.

What should I set Tail Length to?

Start with Learn's estimate and compare exposed phrase endings. Increase it if long tails remain; shorten it when endings, noise floor or direct sound become dull or overprocessed.

Why does RX De-reverb make vocals sound hollow?

It may be treating useful vocal sustain as room. Lower Reduction, revise the frequency profile and Tail Length, and compare the processed voice with the original. Use Output Reverb Only to investigate the loss, not as a rule that every audible word in reverb is damage.

Should I use De-reverb or Dialogue Isolate?

Use De-reverb for general sources and a learned room profile. Use Dialogue Isolate when speech separation from both reverb and changing noise is the central problem.

Should De-reverb come before denoising?

The manual allows either order with Spectral De-noise. Compare restrained treatments on separate copies and choose the result with clearer speech and steadier ambience; neither order is automatically better.

Is two-pass De-reverb better than one pass?

It can be for complex rooms when one light pass targets the long tail and another targets early reflections. A second pass is not automatically better and needs its own bypass test.

Can De-reverb fix a distant microphone?

It can reduce the apparent room and improve focus, but it cannot restore direct detail that the microphone never captured. Severe distance may still require re-recording or ADR.