Quick answer: Voice Isolation separates a recorded human voice from the sound competing with it. Dialogue Leveler does a different job: it rides loud dialogue down and brings soft dialogue up. Voice Isolation requires DaVinci Resolve Studio, while Dialogue Leveler is available on desktop in both Free and Studio. My order: match obvious clip-gain differences first, isolate only when noise masks speech, level second, then EQ or compress only if the mix still needs it.
The two tools sit close together in the Audio Inspector, which is why they get confused, but they solve different problems. If the problem is an air conditioner, a crowd, traffic or room sound, test Voice Isolation, and use the broader Resolve background-noise workflow when hum, broadband noise or another repair needs a different tool.
If the speaker turns away from the microphone, or two people land at different levels, test Dialogue Leveler. And if only two words are wrong, I'd skip both processors: manual gain on those words is usually cleaner.
Voice Isolation vs Dialogue Leveler
| Tool | Problem it solves | Controls | Edition |
|---|---|---|---|
| Voice Isolation | Background ambience, noise, music, or competing sound around a voice | Enable and Amount 0–100% | Studio |
| Dialogue Leveler | Speech that varies too much between phrases or speakers | Preset, Reduce Loud Dialogue, Lift Soft Dialogue, Background Reduction, Output Gain 0 to +6 dB | Free and Studio |
| Clip gain / automation | A few predictable level jumps or words needing editorial judgment | Clip gain, keyframes, fader automation | Free and Studio |
| Compressor | Remaining dynamics needing threshold-, ratio-, and timing-based control | Threshold, ratio, attack, release, makeup gain | Free and Studio |
This page follows the September 2026 DaVinci Resolve 21.1 Reference Manual. The instructions cover desktop Resolve on macOS, Windows and Linux. The manual labels Voice Isolation as Studio-only and documents Dialogue Leveler separately without that restriction.
One boundary up front. Ordinary clip and Track FX instances process existing recordings, not a general live input. The dedicated Cut page Voiceover Tool is a separate recording path that can apply Voice Isolation and Dialog Leveler as it records. That exception doesn't turn either effect into a general live-stream noise filter.
What Voice Isolation Actually Does
Voice Isolation uses an AI model trained on human voices to separate recorded speech from competing sound. The Amount control is a wet/dry blend between the original source and the isolated voice: at 50, the two are roughly equal.
Blackmagic recommends 70–80 as a range that sounds natural while still strongly isolating the source. It's a starting range, not a target every recording has to reach. Push Amount only until the background stops masking the words.
Voice Isolation supports mono, stereo and dual-mono audio tracks, not tracks with more than two channels. Clip controls also depend on the selected source and its channel configuration, so putting a multichannel source on a mono track doesn't by itself prove that its clip controls will be available.
Resolve 21.1 documents Voice Isolation in the Cut page Voiceover Tool during recording. For existing material, keep the original media and an unprocessed timeline copy, so the recorded sound is always there to go back to. Separation can't guarantee usable words when clipping, a siren or another voice covers the speech.
Start by auditioning the hardest line you think is recoverable. Raise Amount only until the competing sound stops masking words, then compare against bypass at matched loudness. Keep enough stable ambience for the cut to sound continuous. If consonants smear, the room tone pumps or the voice turns watery, split the difficult section or reduce Amount instead of treating 70–80 as a target. When Voice Isolation can't separate the line cleanly, RX 12 Dialogue Isolate exposes separate voice, noise and reverb gains for exactly that case.
What Dialogue Leveler Actually Does
Dialogue Leveler analyzes dialogue, rides louder areas down, lifts softer areas, and can lower background it classifies as non-dialogue. It gives you something like detailed clip-gain or fader work without drawing every move yourself. It also keeps more of the original relative dynamics than simply crushing the track with a limiter would.
The controls that actually matter:
- Preset: in the detailed Cut, Edit and Fairlight plugin documentation, Allow wider dynamics is the default, meant for medium-to-high clip levels with wide loud-to-soft changes. More lift for low levels suits quiet dialogue; Lift soft whispery sources addresses whispered dialogue with background noise; Optimize moderate levels suits material at medium levels throughout. Choose by source. It isn't a quality ladder.
- Reduce Loud Dialogue: rides peaks in louder speech downward, a bit like a limiter with no threshold or timing to set.
- Lift Soft Dialogue: an on/off switch that raises low-level dialogue. Background sound occurring at the same time as speech may rise too, so listen for newly exposed breaths or mouth noise.
- Background Reduction: gently lowers non-dialogue sound according to the selected preset. It isn't the same separation process as Voice Isolation.
- Output Gain: adds 0 to +6 dB after leveling. Don't treat it as a target-loudness control.
Dialogue Leveler is designed to retain relative dynamics while increasing intelligibility and average level, but your editorial goal still sets the boundary. A wide shot may need to sound more distant than a close-up, and a whisper shouldn't automatically match a shout.
Like Voice Isolation, it works on existing mono, stereo or dual-mono material. Ordinary clip and Track FX instances can't process a general live audio input, and tracks with more than two channels are unsupported.
The Cut page Voiceover Tool can also apply Dialog Leveler during recording; parts of the manual spell it Dialog rather than Dialogue. I'm not extending the desktop Inspector or Mixer steps on this page to the iPad interface.
Where to Find the Controls
On the Edit or Fairlight page, select the audio clip, open the Inspector, go to Audio and enable the module you need. One warning: the preset documentation is inconsistent. The Cut and Edit Inspector summaries on pages 743 and 928 list Optimized for most sources as the default in a three-preset menu, while the detailed plugin sections on pages 775, 1116 and 4015 list four presets with Allow wider dynamics as default. The descriptions above follow those detailed sections, not a verified universal menu, so check the labels in your installed build and in the clip or track panel you've chosen.
For track-wide treatment, open the Fairlight Mixer. If its Track FX section is hidden, click the Mixer's three-dot menu and select Track FX. The manual's Track FX instructions put Voice Isolation and Dialogue Leveler at fixed, dedicated points, both visible by default once that section is shown; Visible Track FX controls which modules appear. Dialogue Leveler's scrolling processing waveform belongs to the track plugin, not to its clip Inspector version.
Clip processing is the safer choice when only a few shots have noise or level problems. Track processing is convenient when one microphone and one environment remain consistent. I wouldn't apply a track-wide setting to a lane that mixes lavs, booms, phone recordings and room tone unless you've auditioned every transition.
Controls missing? Check the installed edition, select one eligible audio clip and confirm its source channels and track format. Test a separate mono or stereo clip first. Don't unlink a working edit or remap channels blindly; preserve sync and channel assignments.
Best Processing Order When You Need Both
- Listen raw. Mark noise, level jumps, clipping, overlap and unusable words across the entire scene.
- Edit first. Choose the cleaner microphone or take, remove covered words where you can, and split clips when conditions change radically.
- Match gross clip gain. Bring edit-to-edit differences into the same working neighborhood without flattening the performance.
- Apply Voice Isolation only where needed. Raise Amount until the noise stops masking speech, then back off if the voice loses texture.
- Add Dialogue Leveler. Choose the preset matching your source from the labels available in your panel; the detailed manual sections describe Allow wider dynamics as the default. Enable only the modules that solve a problem you can hear. Output Gain isn't a target-loudness control.
- Shape tone. Use the steps in our dialogue EQ guide after restoration, so you're not boosting noise the isolation stage will remove later.
- Control remaining dynamics. Add light compression or automation only if the leveler hasn't already done enough.
- Mix in context. Bring room tone, music and effects back around the cleaned voice.
- Measure the final bus. Use the LUFS and true-peak workflow after all processing is done.
That list is a listening and adjustment sequence. Enabling effects in that order doesn't rearrange their signal path. Voice Isolation and Dialogue Leveler have fixed, dedicated Track FX positions, and the Mixer's Effects Order control, which reorders ordinary inserts, Dynamics and EQ, doesn't move those Track FX slots. Keep clip and track processing distinct so you don't apply the same treatment twice by accident. Track Trim comes after Track FX, as manual page 3940 explains, so neither track Trim nor the track fader should be treated as an input-gain control for these processors.
Clip-Level or Track-Level Processing?
| Scope | Use it when | Risk |
|---|---|---|
| Clip | Noise, mic position, or speaker changes between edits | Different settings can make room tone jump at every cut |
| Track | One consistent microphone, speaker, and environment | A single setting may overprocess exceptional clips |
| Split tracks | Boom, lav, phone, ADR, and narrator need separate treatment | More routing work, but clearer control and automation |
If you split speakers or microphones across tracks, name and color them before adding effects. Keep dialogue, room tone, music and effects on intentional buses; our Fairlight bus-routing tutorial stops a cleaned voice from accidentally feeding the wrong print master.
Compare Bypass at Matched Loudness
Keep the source file unchanged and render tests to a new filename. Toggle the effect's enable/bypass control on the same phrase and compensate output or monitor level until the perceived loudness is close. A louder result can seem clearer even when it carries more artifacts. Listen for consonant edges, sibilance, room tails, breaths and the transitions into silence.
Test four places, not one easy loop: the noisiest sentence, the softest line, a loud line, and an edit where the microphone or room changes. Then render the hardest 20–30 seconds and listen outside Resolve on headphones, on speakers and on a small mono-compatible device.
If playback becomes unreliable after you enable an effect, bypass that instance and compare the same passage. Note the approved settings, re-enable, and check a short export before delivery. No fixed CPU percentage or render-speed gain applies to every computer, source and processing chain.
When Neither Tool Is the Right Fix
- Clipped recording: don't expect these tools to repair recording overload. Try an unclipped microphone or take before pushing isolation harder.
- Covered speech: a horn, clap, or second voice may occupy the same time-frequency region as the word.
- Very distant microphone: reducing room sound won't give you the natural detail of a closer recording.
- A few bad words: clip gain or automation is faster and more transparent.
- Music or effects: these dialogue tools can mistake sung or speech-like material for the target.
- Mixed program master: process dialogue stems before the final mix whenever possible.
For missing or covered words, an alternate take, a different lav/boom choice, ADR, captions or an editorial rewrite is often more honest than extreme restoration. The goal is intelligibility, not pretending the source contains information it doesn't.
Free vs Studio: The Correct Answer
On desktop, Dialogue Leveler works in Free and Studio; Voice Isolation requires Studio. The current Studio overview lists voice isolation among its paid Neural Engine AI audio tools. Blackmagic's September 2026 Studio and iPad feature matrix for 21.1 marks desktop Dialogue Leveler Yes and Voice Isolation Studio on page 8, which matches the edition split in the current manual.
Dialogue Separator is a different Studio tool, with independent dialogue, background and ambience levels. The Resolve 21.1 manual still calls it mono-only: a stereo source has to be split into dual-mono and the two mono channels processed separately. Don't confuse that limit with the mono/stereo support of Voice Isolation and Dialogue Leveler.
Troubleshooting Artifacts
The voice sounds watery or robotic
Reduce Voice Isolation Amount and compare the same difficult section with bypass. Keep enough stable room tone for adjacent edits to match. If the words still sound smeared or metallic, a cleaner take or microphone beats a stronger setting. And don't enable Dialogue Leveler's Background Reduction automatically just because isolation is on; audition whether a second background-reduction stage helps.
Breaths and mouth noise surge
Disable Lift Soft Dialogue for that clip, then correct the quiet phrase or offending breath by hand. It's a switch, not an amount slider. If the added noise occurs with speech, compare a different preset and the isolation setting; don't assume Output Gain can change the balance between voice and noise.
Noise changes at every edit
Use different clip-level settings for different microphones and locations, then bridge the cuts with matching room tone and short crossfades. No single track-wide value can make unrelated rooms identical.
Dialogue Leveler pumps
Match clip gain first, choose a gentler preset, and inspect any dialogue compressor settings or limiters later in the chain. Stacking two aggressive level-control stages can exaggerate the movement.
The last word is cut or fades strangely
Check whether the original edit cuts through a word, a breath or a room tail. Extend handles only where source media exists, then compare with both processors bypassed. If the raw ending is intact but the processed one isn't, back off the processing or use a manual edit; don't assume the problem is missing source audio.
The processed export differs from playback
Render a short WAV test from the correct bus, reimport it, and compare phase, channel layout and level. Our audio export checklist covers the bus and channel-selection mistakes behind most of these mismatches.
Final Dialogue QA
- Listen to every processed edit at normal speed.
- Compare bypass at matched loudness.
- Check breaths, S sounds, plosives, and word endings.
- Listen for ambience opening and closing around speech.
- Confirm intentional distance and performance dynamics remain.
- Check downstream EQ, compression, limiter, and bus routing.
- Render and reimport the hardest passage.
- Measure final loudness and true peak only after the mix is complete.
Voice Isolation should make the background less distracting. Dialogue Leveler should make the performance easier to follow. When the tool itself becomes the first thing you hear, it's doing too much.
Use the DaVinci Resolve Fairlight audio hub to connect cleanup with routing, EQ, compression, export and loudness checks. For licensing decisions beyond this pair of tools, continue with the DaVinci Resolve Free vs Studio comparison.
Voice Isolation vs Dialogue Leveler FAQ
What is the difference between Voice Isolation and Dialogue Leveler?
Voice Isolation separates a recorded human voice from competing sound. Dialogue Leveler rides louder dialogue down, lifts softer dialogue and can reduce some non-dialogue background, so the level stays more consistent.
Is Dialogue Leveler available in DaVinci Resolve Free?
Yes. On desktop, Dialogue Leveler is available in both Resolve Free and Studio. Voice Isolation requires Resolve Studio.
Which tool should go first?
When you need both, match large clip-gain differences first, apply Voice Isolation conservatively, then add Dialogue Leveler. Treat that order as a starting point and compare the full chain at matched loudness.
Can Voice Isolation remove wind and crowd noise?
It can reduce them as long as the speech remains recoverable. It can't reliably reconstruct words covered by strong gusts, clipping, applause or another voice.
Why does Dialogue Leveler raise breaths and room noise?
Lift Soft Dialogue raises quiet dialogue, and noise occurring with speech may rise with it. Disable that switch or correct the phrase manually. Output Gain changes the level after processing; it isn't a noise-reduction amount or a loudness target.
Should I still use a compressor?
Only if the dialogue still needs extra controlled dynamics after clip gain and leveling. Start gently and check that the compressor isn't exaggerating pumping or breaths.
Can I apply these tools to an entire track?
Yes, on existing mono, stereo, or dual-mono tracks. Ordinary clip and Track FX instances can't process a general live audio input. The Cut page Voiceover Tool is a separate recording workflow that can apply Voice Isolation and Dialog Leveler while it records. Track processing is safest when the microphone, speaker, and room stay consistent.
