iZotope RX: The Complete Guide for Music Producers

When You Actually Need RX

The most useful starting point is understanding the specific scenarios where RX provides value that no other tool can match. If none of these apply to your work, you may not need RX — Waves Clarity Vx at $49 handles straightforward vocal noise reduction, and your DAW's built-in noise reduction handles basic applications.

RX is genuinely necessary when: you record vocals in acoustically imperfect environments and standard noise reduction produces artifacts; you have recorded a great performance with one specific problem (a phone ring, a cough, a door slam) that you cannot re-record; your audio has clipped at recording and the distortion is audible; you need to separate stems from a fully mixed recording; you are editing dialogue for video and need advanced noise control; you work with archival or found audio that has accumulated noise, hiss, and artifacts over time.

If your work is primarily software-instrument-based production with occasional clean vocal recording in a quiet environment, RX Elements at $99 covers everything you realistically need. The Standard and Advanced tiers earn their cost in specific professional contexts.

Voice De-noise — The Starting Point

Voice De-noise is the RX module that most music producers reach for first, and for good reason — it is the most immediately applicable to the common problem of home studio vocal recording.

Voice De-noise uses AI to separate the vocal signal from background noise — the same source separation approach as Waves Clarity Vx, but integrated into RX's broader restoration framework. The key parameters:

Threshold: The aggressiveness of noise reduction. Higher values remove more noise but increase the risk of artifacts on the voice. Starting point: 40-50%. Increase until noise is reduced to acceptable levels, stop before artifacts appear on the vocal.

Voice activity detection: RX identifies which portions of the audio contain voice and applies noise reduction more aggressively to silent sections. This produces cleaner silence between phrases without over-processing the vocal itself. Keep this enabled in standard use.

Mode (Standard vs Extreme): Standard mode is appropriate for noise at moderate levels. Extreme mode applies more aggressive processing for severe noise situations — voices recorded in very loud environments. Extreme mode produces more artifacts at equivalent noise reduction amounts; use Standard unless Standard cannot achieve sufficient noise reduction.

Practical workflow: insert Voice De-noise as a plugin on the vocal track, or send the audio to RX standalone via RX Connect. Set Threshold to 40%, enable voice activity detection, play the audio, and adjust Threshold upward until the noise between vocal phrases is at an acceptable level. Do not aim for absolute silence — a small amount of residual noise is preferable to artifacts on the voice.

Spectral Repair — Surgical Audio Removal

Spectral Repair is the most powerful and unique capability in RX — the module that can fix audio problems that were previously impossible to address without re-recording. It requires RX Standard or Advanced.

The spectrogram display — a visualization of audio showing frequency content over time as a heat map — allows you to see individual sounds as visual objects. A phone ring appears as a cluster of energy at specific frequencies at a specific point in time. A cough appears as a broadband burst of energy. A door slam shows as a specific transient cluster. A low-frequency hum shows as a horizontal line at 60Hz (or 50Hz in non-US countries) running continuously through the recording.

The repair process: identify the problem visually in the spectrogram, draw a selection around it using the selection tools (lasso for irregular shapes, rectangle for regular ones), choose a repair algorithm, and apply. RX interpolates what the audio should have sounded like based on the surrounding material and reconstructs the selected region.

Repair algorithms:

Interpolate: Reconstructs the selected region from the audio immediately before and after it. Best for short, isolated sounds — a single phone ring, a brief click, a momentary noise. The interpolation only works convincingly for durations up to about 0.5–1 second on most material.

Replace: Replaces the selected region with audio from elsewhere in the recording — similar ambient content sampled from a nearby region. Best for room tone and noise floor replacement when interpolation cannot produce convincing results for longer selections.

Pattern: Analyzes the surrounding audio for repeating patterns and uses them to reconstruct the selected region. Best for tonal and musical content where the interpolated material needs to maintain a melodic or harmonic relationship with surrounding audio.

Attenuate: Reduces the level of the selected region rather than removing it entirely. Best when full removal creates an obvious gap — attenuating a cough to reduce its prominence rather than eliminating it entirely can produce more natural results.

De-clip — Recovering Clipped Audio

Clipping occurs when a recorded signal exceeds 0dBFS — the maximum level a digital recording can represent. The peaks of the waveform are cut off, producing harsh distortion that is immediately audible as a crackling, buzzing, or crunching quality on the affected audio.

De-clip analyzes the clipped waveform and uses algorithmic extrapolation to reconstruct what the waveform peaks should look like based on the surrounding signal shape. The algorithm cannot perfectly reconstruct heavily clipped audio — if 20% of the waveform is clipped, significant reconstruction quality is lost. But for moderately clipped audio (occasional peak clipping up to about 6dB over 0dBFS), De-clip can reduce the distortion from prominently audible to acceptable or inaudible.

Clipping threshold: RX automatically detects the clipping level, but you can set it manually if the automatic detection is off. Set it to where the clipping begins — usually 0dBFS for digital clipping.

Algorithm: Clipfix (standard) works well for most material. Declip 2 is a more computationally intensive option that produces better results on sustained tonal material like sustained notes and vocals. Use Declip 2 when standard Clipfix produces remaining artifacts on tonal content.

Output gain: De-clipped audio will be slightly lower in level than the original because the reconstructed peaks are correctly lower than the clipped values. A small amount of output gain compensation (1–3dB typically) returns the level to approximately its original perceived loudness.

Music Rebalance — AI Stem Adjustment

Music Rebalance uses AI source separation to identify and adjust the relative levels of stems within a fully mixed recording — vocals, bass, drums, and other instruments. This capability is remarkable for what it allows: modifying the balance of a finished mix without access to the original project files or stems.

Practical applications: Creating an instrumental version of a fully mixed song by reducing the vocal level. Boosting the bass in a mixed recording that was mastered too thin. Reducing overpowering drums in a live recording where the balance cannot be corrected in the original session. Remixing archival material that exists only as finished mixes. Creating stems from finished mixes for re-release or licensing purposes.

Quality limitations: Music Rebalance's source separation is imperfect — complete separation of tightly integrated mixes produces artifacts (remnants of other elements in the separated stems, frequency smearing at transition points). The quality is sufficient for many practical applications but does not match the quality of properly recorded individual stems. For creative and practical applications where perfect transparency is not required, the results are impressive. For critical professional applications where artifact-free stems are essential, there is no substitute for access to the original session.

Settings: Each stem (vocals, bass, drums, other) has a gain slider ranging from -inf (silent) to +6dB (boost). Adjust gradually — small changes produce cleaner results than extreme adjustments. The sensitivity slider affects how aggressively the algorithm pursues source separation; higher sensitivity can produce more complete separation but also more artifacts.

RX Connect — DAW Integration

RX Connect is the bridge between your DAW and the RX standalone application. Without RX Connect, the workflow requires exporting audio from your DAW, opening it in RX standalone, processing it, exporting again, and re-importing — a workflow that is functional but slow when you need to iterate on multiple clips in a session.

With RX Connect: insert the RX Connect plugin on a DAW track. Select the region of audio you want to process. Click "Send to RX" in the plugin. The audio opens automatically in RX standalone. Apply processing. Click "Send Back" in RX. The processed audio returns to the exact position in your DAW. The round trip takes seconds rather than minutes.

RX Connect is available as a plugin for Ableton Live, Logic Pro, Pro Tools, FL Studio, Cubase, and other major DAWs. It is included with RX Standard and Advanced, and available as a separate purchase for RX Elements users.

De-hum — Electrical Noise Removal

Electrical hum — the 50Hz or 60Hz tone (and its harmonics) introduced by ground loops, poorly shielded cables, or nearby electrical equipment — is one of the most common recording problems. De-hum addresses it with a comb filter that precisely removes the fundamental frequency and its harmonics without affecting the rest of the audio.

Frequency: Set to 60Hz for North American electrical systems, 50Hz for European and most other regions. If you are uncertain, zoom in on the spectrogram display — electrical hum appears as a series of horizontal lines at 60Hz, 120Hz, 180Hz, 240Hz (or 50Hz, 100Hz, 150Hz, 200Hz for 50Hz systems).

Number of harmonics: Electrical hum generates harmonics — multiples of the fundamental frequency — that need to be removed along with the fundamental. Start with 10 harmonics and adjust. More harmonics removes more of the hum character but can begin affecting content at higher harmonics.

When De-hum is not the right tool: Broadband noise, inconsistent noise, and noise that is not electrical hum in character (traffic, HVAC, computer fans) should be addressed with Voice De-noise or the Dialogue De-noise module rather than De-hum. De-hum is specifically for tonal, constant-frequency electrical noise.

Practical Workflow for Music Producers

The most efficient RX workflow for music producers who primarily work in a DAW:

1. Record your audio normally. Address obvious technical problems (gain staging, microphone placement) at the recording stage — RX fixes problems but cannot restore quality that was never captured.

2. On the first listen after recording, note any specific problems — a noise in the background, a cough, a clicky consonant, a section that feels noisy.

3. Use RX Connect to send the affected region to RX standalone. Open the spectrogram display. Identify the problem visually.

4. Apply the appropriate module — Voice De-noise for background noise, Spectral Repair for specific sounds, De-clip for overloaded recordings, De-hum for electrical noise.

5. Preview the result in RX before sending back. Toggle between the processed and original audio using the A/B comparison.

6. Send the processed audio back to your DAW via RX Connect.

7. Continue the normal mixing workflow with the repaired audio.