> ## Content Index
> Fetch the complete content index at: https://izotoperx.net/llms.txt
> Use this file to discover other available public pages before exploring further.

# iZotope RX 12 Repair Assistant: Complete Guide
- URL: https://izotoperx.net/izotope-rx-repair-assistant/
- Published: 2026-09-03T14:29:40.000Z
- Updated: 2026-09-05T07:27:53.000Z
- Description: Let Repair Assistant diagnose the recording, compare its suggestions at matched loudness, then reduce or bypass the blocks that solve the wrong problem.
- Author: Brandon Hayes
- Tags: Getting Started, RX 12

To use iZotope RX 12 Repair Assistant, work on a copy, choose Voice for speech or Music for music and sung vocals, and click Learn on a representative problem passage. In a compatible plug-in host, feed it playback until learning finishes. Then compare the suggested processing with bypass, reduce any block that damages the wanted sound, and check the complete exported result.

Repair Assistant is a useful first diagnosis when several problems arrive together: hiss behind a voice, excessive room sound and sharp sibilance, for example. Its suggestion still needs judgment. A quieter recording is not an improvement if words become watery or instruments lose their attack. The controls below follow the RX 12 manual checked on September 5, 2026.

## What RX 12 Repair Assistant actually does

The assistant detects common recording problems and proposes settings you can refine. It is included in all three editions, but the workflow depends on your purchase: Elements supplies the plug-in; Standard and Advanced also include the standalone RX Audio Editor. The [current RX edition comparison](https://www.izotope.com/products/rx-advanced?tab=compare&ref=izotoperx.net) lists those differences and the improved RX 12 Repair Assistant neural networks.

It does not repair an abstract quality called “bad audio.” It estimates which problems are present and exposes controls for a collection of processes. That can shorten the search for a useful repair, but it cannot decide whether a breath, cymbal wash, room decay or pick attack belongs in the performance.

There is also a documentation mismatch to know about. The [RX 12 product page](https://www.izotope.com/products/rx-elements?ref=izotoperx.net) advertises four source categories and Light, Medium and Aggressive variations. The [RX 12 Repair Assistant manual](https://docs.izotope.com/rx12/en/repair-assistant.html?ref=izotoperx.net) instead documents Voice and Music with adjustable processing sections. This guide follows that manual; the advertised intensity labels are not required steps in the workflow below.

## Choose the right source type before analysis

| Documented mode | Starting material                                           | Listen for after learning                           |
| --------------- | ----------------------------------------------------------- | --------------------------------------------------- |
| Voice           | Spoken dialogue, podcasts and voice-over                    | Consonants, breaths, room tone and sibilance        |
| Music           | Full mixes, individual instruments and isolated sung vocals | Note decay, harmonics, transients and vocal texture |

Start with the mode that matches the wanted material. The manual places isolated singing under Music. Voice may still be worth a deliberate comparison when you want its speech-oriented controls, but that is an experiment, not the documented default for every vocal.

A mixed program needs extra care. If you have separate dialogue and music tracks, repair the dialogue before combining them. On a finished mix, listen to both layers while changing a control. A cleaner voice is not a useful trade if the background music starts pumping. For pure sound effects, the current manual does not specify a separate Sound Effects mode; compare a targeted tool if neither documented mode gives a natural result.

## Module versus plug-in workflow

In the Standard or Advanced Audio Editor, disable [Instant Process](https://docs.izotope.com/rx12/en/interactive-tools.html?ref=izotoperx.net#instant-process) before making a selection, then open Repair Assistant from the upper-right button. Choose Voice or Music and click Learn. Include wanted signal and the defect: unlike noise-profile learning, this analysis does not require a noise-only segment. With no selection, the manual says it analyzes as much audio as needed; it does not promise to analyze every second of the file.

For a plug-in, choose the mode, click Learn, then play the track until the assistant finishes listening. Learning needs audio to reach that instance. A stopped transport, muted route or clean lead-in can explain why the result is missing or unrepresentative.

The [RX 12 release notes](https://www.izotope.com/pages/release-notes/rx-standard?ref=izotoperx.net) list Repair Assistant specifically as AU, VST3 and AAX AudioSuite. Do not infer real-time AAX insert support from the bundle's broader format list. In Pro Tools, use the AudioSuite workflow.

**If AudioSuite learning appears stuck:** the Repair Assistant manual requires a clip longer than seven seconds and warns that shorter clips can learn indefinitely. Test a longer representative clip and feed the AudioSuite preview to the processor. The manual also mentions looping under AAX, but the current format list does not establish a real-time AAX Repair Assistant insert; a longer AudioSuite clip is the documented route to start with.

Our [Editor, plug-in and Connect workflow guide](https://izotoperx.net/rx-editor-plugins-connect-workflow/) explains which route suits a selection-based repair or an ongoing DAW session. Elements users should follow the plug-in branch rather than look for an included Editor.

## Nine steps for a useful Repair Assistant pass

1. **Preserve the source.** Keep an untouched audio-file copy outside the working edit. Duplicating a clip or playlist may still reference the same media.
2. **Disable Instant Process and select.** In the Editor, turn Instant Process off before selecting a representative passage with wanted audio, the defect and a transition. Include the intended channels and full frequency range.
3. **Choose Voice or Music.** Start with Voice for spoken dialogue and Music for music, instruments or isolated singing.
4. **Run Learn.** In the Editor, analyze that selection. In a plug-in, click Learn and feed playback until learning finishes; use a clip longer than seven seconds in AudioSuite.
5. **Compare with the original.** Listen to the suggested result and bypass at similar perceived loudness on the same passage.
6. **Reduce unnecessary processing.** Back off the individual controls where the repair harms words, note decays or attacks. A named intensity preset is not required.
7. **Inspect each section.** Use the section On/Off controls to compare Clean Up, Tone, De-ess or De-harsh, and De-clip individually.
8. **Check the removed signal.** Use the section Audition control to hear what it removes, then switch back to normal output and judge a second difficult passage.
9. **Apply once and export.** Undo test renders or reopen the untouched working audio before processing the intended range once. In the Editor, Render and export to a new filename; in a plug-in workflow, use the host's render or bounce. Listen through the complete output in context.

A ten-to-twenty-second test passage is a practical starting point, not a universal RX requirement. The separate AudioSuite limit above still applies. Choose an excerpt with the quietest wanted detail and a difficult interruption, then test another acoustic condition before committing a long program.

## How to compare suggestions fairly

Start with the suggestion that preserves the source, not the one that produces the darkest noise floor. Match the processed and bypassed output by ear. A small level lift can make a result seem more detailed even when consonants or transients are worse.

Listen in three passes. First, judge intelligibility or musical intent. Second, focus on the defect: hum, hiss, clipping, harshness or reverb. Third, ignore the defect and listen only for new damage—flutter, hollow tone, softened attacks, unstable ambience or stereo movement.

Keep the audition passage fixed while changing settings. If you move to a different phrase each time, source variation hides the processing difference. In the Editor, use the [Compare Settings and Preview controls](https://docs.izotope.com/rx12/en/common-module-controls.html?ref=izotoperx.net#compare) where available; Render applies processing. RX plug-ins do not have the Editor module footer controls apart from Bypass, so use host snapshots or duplicate working tracks for alternative settings. Never leave both comparison tracks audible together.

## What happened to Light, Medium and Aggressive?

If you arrived from a tutorial or product description promising Light, Medium and Aggressive, do not spend the session searching for those buttons in a different interface. The current manual's workflow is Learn followed by adjustments to individual sections. Treat intensity as how much repair you apply, rather than a compulsory three-step ladder.

Begin with the learned suggestion, then reduce or bypass the section that changes wanted audio. Make one adjustment, listen to the same line and compare again. If stronger de-noising dulls consonants, restore some noise; if tonal shaping makes adjacent clips sound different, reduce it or leave Tone off.

The delivery context matters. A documentary line under music can tolerate residue that sounds obvious in solo. A dry audiobook exposes lisping and unstable room tone. Approve the lightest processing that solves the distraction where the listener will hear it.

## Read the Voice blocks one by one

**Clean Up:** In Voice mode, the De-noise control applies Dialogue Isolation processing, while De-reverb controls room reduction. De-hum operates only when hum was detected during learning; De-click is another toggle. If a hum switch seems ineffective, recheck the learned passage rather than repeatedly raising unrelated noise reduction.

**Tone:** Tilt shifts the balance between low and high frequencies; Scoop changes the relationship between the midrange and the extremes. Lo-cut and Hi-cut are separate switches. Compare adjacent clips before accepting a broad tonal change. A more polished solo voice can sound conspicuous when it no longer matches the interview.

**De-ess:** Threshold determines when sibilance processing starts; Shaping changes the sibilant's spectral shape. Zero Shaping does not mean the whole De-ess section is bypassed. Use its On/Off switch for that comparison and listen for lisping or missing consonants.

**De-clip:** set the detection threshold against the clipped concentration in its histogram, then check Output Gain and headroom. The Limiter is optional true-peak protection after processing, not proof that the distortion is repaired. For a waveform with flattened peaks, the dedicated [RX De-clip workflow](https://izotoperx.net/fix-clipped-audio-rx-de-clip/) gives the longer diagnostic sequence.

## Read the Music blocks without sanding off the performance

Music replaces De-ess with De-harsh. Its Clean Up section uses adaptive de-noising plus De-squeak, De-hum and De-Pick controls. De-hum still depends on detection during learning. The dedicated De-harsh cutoff marks the lower boundary for harshness detection; do not treat it as a low-pass filter that should remove every bright sound.

Attack and decay are the danger zones. A brush snare, finger-picked guitar or bright piano can look like unwanted high-frequency energy to a broad process. Use the Ear monitor and bypass to check whether the assistant is removing the nuisance or rewriting the musician's articulation.

If the problem is one narrow event, stop using a track-wide macro. A single chair squeak, cough or dropped object belongs in a local spectral selection. The [RX spectrogram tutorial](https://izotoperx.net/how-to-read-izotope-rx-spectrogram/) shows how harmonic lines, transients and broadband events differ visually.

## Use the Ear control as a truth test

The manual calls the ear-style control **Audition**: each section can output what that section removes. You want to hear the defect. If the monitor contains recognizable words, clean guitar harmonics, cymbal attacks or the body of a snare, compare a weaker setting and judge the normal output again.

Do not demand complete silence in the removed channel. Separation and denoising often include faint traces of the wanted source. The practical question is whether backing off improves the final output more than it restores the defect.

Alternate between the removed signal, normal processed output and bypass. Turn Audition off before the final render or bounce. Nick Messitte's [RX 11 Repair Assistant demonstration](https://www.izotope.com/community/blog/repair-assistant?ref=izotoperx.net) illustrates the trade-off: his adjusted clipping example improves one aspect while adding an audible word-ending artifact. It is a useful listening exercise from an older version, not an RX 12 performance benchmark.

## When to leave Repair Assistant and open a dedicated module

| What you hear or see                     | Better next tool | Why                                                  |
| ---------------------------------------- | ---------------- | ---------------------------------------------------- |
| Steady 50/60 Hz tone and harmonic ladder | De-hum           | Targets a known tonal interference pattern           |
| Individual clicks or digital impulses    | De-click         | Exposes sensitivity and click-specific controls      |
| Clearly flattened peaks                  | De-clip          | Lets you inspect the threshold and repaired waveform |
| One isolated visual intrusion            | Spectral Repair  | Limits processing in time and frequency              |
| Changing noise around spoken dialogue    | Dialogue Isolate | Separates voice, noise and reverb directly           |
| Ordinary steady voice noise              | Voice De-noise   | Offers a focused, lighter speech-denoise workflow    |

The dedicated tools in the table depend on your edition. Elements includes De-hum, De-click, De-clip and Voice De-noise plug-ins; Spectral Repair and Dialogue Isolate require Standard or Advanced. For changing speech backgrounds, compare the [Dialogue Isolate tutorial](https://izotoperx.net/izotope-rx-dialogue-isolate/); for clicks and pops, use the [De-click guide](https://izotoperx.net/izotope-rx-de-click-tutorial/).

## Long recordings need representative passes

An hour-long interview rarely has one acoustic condition. Air conditioning changes, the speaker leans away, a second microphone opens and the room grows louder. One analysis over the complete file can encourage a compromise that is ideal for none of those sections.

Divide the program by recording condition, not arbitrary duration. Analyze a representative bad passage from each scene, microphone or room. Keep the processing consistent within a scene, but do not force the same controls onto a different source.

Do not set a deadline from somebody else's learning time on a different machine or RX version. Short excerpts let you reject a poor setting before committing the program. If the assistant cannot finish on a sensible excerpt, check signal flow, the AudioSuite clip limit and host compatibility before assuming the recording needs more aggressive processing.

## Why the result sounds worse

**The source mode is wrong.** Re-analyze with the category that best matches the wanted signal.

**The test section is unrepresentative.** Include the actual defect, wanted audio and a transition instead of learning from a clean lead-in.

**Processing strength is hiding a bad choice.** Reduce the individual controls and identify the section that damages the source.

**Several blocks solve the same symptom.** Too much Clean Up plus tonal shaping can make the source hollow. Bypass one process at a time.

**You are judging in solo.** Put the repaired audio back under music, effects or adjacent dialogue and match level. Some residual noise is less objectionable than an exposed processing artifact.

**The recording is beyond reconstruction.** Repair can reduce a defect; it cannot recover words or harmonics the microphone never captured. Consider an edit, alternate take, ambience cover or re-recording.

## A final render checklist

- The processed and bypassed versions are compared at similar loudness.
- Words, breaths, attacks and note decays remain natural.
- The removed-signal monitor does not carry substantial wanted content.
- No block remains active merely because the assistant selected it.
- Room tone and noise do not jump at edit boundaries.
- The setting has been tested on more than one difficult passage.
- The full render retains sync, channel layout and headroom.
- The repaired file still works on headphones and ordinary speakers.

In Advanced, **Open Module Chain** opens the assistant's recommendations for deeper editing or saving as a chain preset. This particular handoff is Advanced-only; it is not the same claim as saying Standard lacks Module Chain itself. In the Editor's Assistant Preferences, the gear lets you exclude processes from analysis. Those preferences do not apply to the plug-in.

For delivery, [File > Export](https://docs.izotope.com/rx12/en/working-with-files.html?ref=izotoperx.net#export-file) writes an audio file such as WAV or AIFF. Ordinary Save overwrites an open WAV or AIFF; Save RX Document keeps the editable RX history but produces an .rxdoc for RX, not a normal DAW deliverable. Reopen the exported audio and confirm the intended duration, channels, sync and unclipped peaks.

The [RX 12 beginner tutorial](https://izotoperx.net/izotope-rx-12-tutorial-beginners/) covers the wider Editor workflow, while the [Learn iZotope RX hub](https://izotoperx.net/learn-izotope-rx/) organizes the next skills. Repair Assistant should make your first decision quicker. Your ears still make the final one.

## Frequently asked questions

### Is Repair Assistant included in RX 12 Elements?

Yes, as a plug-in. RX 12 Standard and Advanced also include the Audio Editor and its Repair Assistant module. Elements does not include the standalone Editor.

### Which source type should I choose for a podcast?

Choose Voice. Analyze a passage that contains speech, the main noise and a transition into room tone rather than an empty intro.

### Should I use Light, Medium or Aggressive?

The RX 12 product page uses those labels, but the current manual describes Voice and Music with individual processing controls. Follow Learn, then reduce or bypass damaging sections; a named intensity preset is not required.

### Does Repair Assistant work as a plug-in?

Yes. The RX 12 format list specifies AU, VST3 and AAX AudioSuite for Repair Assistant. Click Learn and feed playback until it finishes. AudioSuite requires a clip longer than seven seconds.

### What does the Ear button do?

Audition monitors what a section removes. If clear words, musical harmonics or important attacks appear there, compare weaker processing or bypass. Turn Audition off before rendering or bouncing.

### Can Repair Assistant replace manual RX modules?

It can produce a useful starting point, but targeted modules offer more control for hum, clicks, clipping, isolated spectral events and difficult dialogue. Use the assistant to diagnose, then go manual when necessary.

### Why does the result sound watery or dull?

Processing may be too strong, the mode may not suit the source, or several sections may be altering the same wanted detail. Reduce individual controls and bypass sections one at a time while comparing the same original passage.

### Should I analyze an entire long recording?

Start with a short representative problem section. For a long program, split the work by room, microphone or recording condition and verify each setting on several difficult passages.