Separating dialogue, music and noise: How to rescue the audio from finished videos

The client signed off on the corporate film a year ago, and now the license for the background music is expiring. The individual tracks from the shoot were deleted long ago, and all that's left is the finished export, in which voice, music and street noise are baked into a single audio track. For a long time, your only options were to redo the audio or keep the music.

Nowadays, AI models can break such a mix back down into its components. Since the update of November 2026, Premiere splits the audio of a clip into dialogue, music, ambience and sound effects. DaVinci Resolve Studio, Logic Pro and iZotope RX have been able to do parts of this for longer, and since November 3 there has been a dedicated app for film audio, Photon from iZotope.

This guide shows which program delivers which tracks, how to proceed in Premiere, DaVinci Resolve and Final Cut Pro, and how you can tell that the separation is reaching its limits.

What happens when an AI breaks down the finished audio?

A trained model recognizes the patterns of speech, music and noises in the audio mix and writes each source to its own track. In audio editing, these partial tracks are called stems. Afterwards, you remix them: the music quieter or muted, the voice louder, the room reverb weaker.

This is not a true undo: the model reconstructs what it believes to be voice or music.

Typical occasions are an expiring music license, a new voice-over in another language, and interviews in which traffic or an audience drowns out the voice. iZotope itself names the first two as use cases for Photon.

Noise that arises during recording is a different topic; basics on it can be found in the article Getting rid of reverb and noise in audio recordings.

Which program separates dialogue, music and noises?

For the audio of finished videos, Premiere, DaVinci Resolve Studio, iZotope RX 12 Advanced and Photon separate into dialogue, music and noises. Logic Pro, Audacity and the Music Remixer in Resolve, on the other hand, are built for songs: they break music down into vocals and instruments.

The free DaVinci Resolve lacks Voice Isolation, Dialogue Separator and Music Remixer. The feature overview for version 21 lists all three as Studio features.

Final Cut Pro does not separate into tracks. Apple describes Voice Isolation as favoring detected human voices over the other parts of the signal. For an interview with street noise this may be enough, but not for replacing the music.

In RX 12 separates the Scene Rebalance module a soundtrack into voice, music and effects, but only in the Advanced edition. Photon runs as a standalone app entirely on your computer and needs the graphics card for that; on the Mac, iZotope requires macOS 14 and an Apple chip from M1 onward.

Among the music tools, the stem splitting in Logic Pro requires a Mac with an Apple chip. For Audacity, the separation is only available via the OpenVINO plug-ins, and according to the download page, these do not yet work with the latest Audacity version.

Cubase Pro has also been splitting recordings into up to six tracks since version 16.

How do you separate the audio in Premiere?

In Premiere, you select the clip, open the Audio section in the Properties panel, and click Enhance Audio. Premiere analyzes the clip and separates the audio into dialogue, music, ambience, and sound effects. The separation itself runs on your device, as the Adobe Help explicitly states.

1. Select an audio or video clip in the timeline.
2. In the Properties panel, open the Audio section and choose Enhance Audio. You can also find the command in the context menu and in the Clip menu.
3. For the dialogue, set the dialogue volume, optimization, and reverb. At 0 percent optimization, the voice retains its original sound.
4. Adjust music, ambience, and sound effects with a volume slider each, or mute a layer completely.
5. Click Split into clips if you want to edit the layers individually or hand them over to the person doing the sound mix.

Enhance Speech V1 runs entirely on the device and costs no credits, while V2 computes in the cloud as part of Firefly and uses two Generative Credits per minute. You switch the model per clip in the Enhance Audio settings. The big advantage over a standalone app: you stay in the editing program, and the separated clips sit in sync in the same timeline.

Premiere analyzes five seconds beyond the clip on both sides. Within this lead-in and lead-out, which Adobe calls handles, you can trim without a new analysis; beyond that, Premiere only computes the new section.

The reverb reduction only affects the speech, and how strong it is depends on the recording, according to Adobe's own description.

How do you proceed in DaVinci Resolve and Final Cut Pro?

In DaVinci Resolve Studio, Dialogue Separator and Music Remixer are located as track effects (Track FX) in the mixer on the Fairlight page. They appear there in the list after the two speech tools Voice Isolation and Dialogue Leveler, but out of the box Resolve only shows those two. This is how Blackmagic describes it in the manual for the new features of version 19, with which the tools were added.

DaVinci Resolve Studio

1. Open the Fairlight page and, on the right in the mixer, the options menu with the three dots.
2. Turn on the Track FX area and, at the bottom in the submenu Visible Track FX (visible track effects), choose the Dialogue Separator or the Music Remixer.
3. Split a stereo track into two mono tracks beforehand, because the Dialogue Separator currently only works with mono.
4. Adjust dialogue, background, and room portion individually and capture changes over time with automation.

Blackmagic gives the example of a person walking toward the camera in a reverberant room: with automation, they sound spacious at first and then increasingly dry, without the background changing. How to work with tracks and the Fairlight mixer in Resolve is shown in the practical course on DaVinci Resolve.

Final Cut Pro

In Final Cut Pro, Voice Isolation does not affect the entire clip, but a single audio component. For a clip with several components, such as dual mono, you therefore select one of them first. The steps apply to Final Cut Pro 12.4 from September 29, 2026.

1. Open the Inspector and click the Audio button at the top.
2. In the Audio Enhancements section, check the Voice Isolation box.
3. In the Voice Isolation section, drag the Amount slider until the voice is clearly in front.

Apple warns that Voice Isolation running on several components at the same time can distort the sound; in that case, you turn it off again for individual components.

Where does the separation reach its limits?

The manufacturers name the weak points in their own help pages: artifacts, remnants of the other sources, and distorted sound.

• Artifacts: In the RX module Dialogue Isolate, which separates the voice from reverb and noise, a higher sensitivity can, according to iZotope, produce artifacts and even make the speech less clear.
• Bleed: In the RX module Music Rebalance, which splits music into voice, bass, drums and the rest, portions of other elements can end up in a track at higher sensitivity.
• Vocal remnants: The Audacity help states that removing vocals often leaves remnants and artifacts behind.
• Distortion: Final Cut Pro can distort the sound if Voice Isolation is active on several components running at the same time.
• Mono: The Dialogue Separator in Resolve only processes mono tracks.

It's best to listen to the result with headphones at points where music and voice play at the same time, and compare it with the original. For Premiere, Adobe advises editing with the local model V1 and testing V2 only for the final mix.

What this means for you

For most rescue cases you no longer need a specialized program: in Premiere, the separation has been part of the editing program since November, while in DaVinci Resolve you need the Studio version for it.

1. Back up the original: Before the separation, export an unchanged copy of the audio track, because the separated tracks are a reconstruction and not a replacement for the original recording.
2. Choose the tool by task: For film sound with voice, music and noises, you use Premiere, Resolve Studio, RX 12 Advanced or Photon; for a song, Logic Pro, Audacity or the Music Remixer.
3. Save credits: In Premiere, work with Enhance Speech V1 and switch on V2 only for the finished version, because each minute there costs two Generative Credits.
4. Swap audio without re-rendering: If the picture stays the same, you only replace the audio track; how this works with two free programs is shown in a lesson on replacing the audio track.

Our assessment: for an interview with music playing too loudly underneath, it is almost always worth trying, because you only have to lower the music and small remnants are less noticeable under the louder voice. The music has to disappear completely, on the other hand, when its license expires; then you hear every remnant, and you should check the result especially thoroughly. For the same reason, you'd better not firmly plan on a song from which the voice is supposed to vanish without a trace.

A clean recording takes the least work; how to prevent reverb and noise from arising in the first place is shown in the course on audio recordings with reverb, noise, and interference sounds.

Information as of: November 15, 2026.

Sources

• Adobe Help: Enhance Audio, Overview
• Adobe Help: Applying "Enhance Audio" to Clips
• Blackmagic Design: DaVinci Resolve 19 New Features Guide
• Blackmagic Design: DaVinci Resolve Studio 21 Features
• Apple Support: Improve Audio in Final Cut Pro for Mac
• Apple Support: Stem Splitting in Logic Pro for Mac
• Apple Support: Features Not Available on Macs with an Intel Chip
• iZotope RX 12 Documentation: Scene Rebalance
• iZotope RX 12 Documentation: Dialogue Isolate
• iZotope RX 12 Documentation: Music Rebalance
• Audacity Support: Isolating or removing vocals from a song