Skip to main content
Community Manager
July 27, 2026
StickyNews

Now in Beta: Separate Crosstalk in Premiere

  • July 27, 2026
  • 6 replies
  • 850 views

Hey everyone! We're excited to share a new Firefly AI-powered audio feature that just landed in Premiere Beta [26.5.0 (Build 75)]: Separate Crosstalk.

If you've ever shot a scene on a single mic – whether that's a boom, a phone, or just one lavalier on set – you know the moment: two people start talking at the same time, the mic can't tell the voices apart, and the only real fix is a re-record. Now Separate Crosstalk gives you a way to pull those overlapping voices apart, right inside your timeline!
 

What's New?

Separate Crosstalk is built for one specific moment: when two speakers talk over each other at the same time on a single recording – actual overlapping speech, not just two people who happen to be in the same scene. It works on mono and stereo clips and separates up to two overlapping speakers from that shared audio. This isn't full diarization and it isn't meant to split every speaker onto their own track across a whole project. It's meant to solve the crosstalk moment that today has no fix in post beyond rerecording.

Just feed it a clip and it returns three tracks: Speaker 1, Speaker 2, and a background/ambience track, so you have full control in the mix.

This is a premium feature. Processing costs 1 Generative Credit per second of audio processed, so it's worth trimming your selection down to the crosstalk section rather than running it on a long clip. Keep in mind that Separate Crosstalk automatically adds a 2 second handle to each side and that added time will be reflected in the cost shown before you process.
 

Where to Find Separate Crosstalk

The workflow starts with finding where the crosstalk actually happens in your clip. Select the clip on your timeline, listen for the section where the two speakers overlap, and either:

  • Trim to that range and process just the overlapping section for the fastest, most targeted result, or
  • Select the full clip and let Separate Crosstalk find the overlap range on its own within it

From there, right-click the clip and choose Separate Crosstalk, or use the quick action in the Properties panel labeled Separate Crosstalk. Both methods will open a dialog box to process the selection. Because it's a premium feature, narrowing to just the crosstalk moment before you process keeps your credit usage efficient.

Clips can run anywhere from 1 second up to 10 minutes. If your overlapping section falls outside that range, trim it down first.


                      

Right-click menu

 

Properties panel quick action
Separate Crosstalk

For editors working in documentary, run-and-gun, or single-mic setups, overlapping dialogue has always been a dead end. Once two voices are captured on top of each other on the same mic, there's no manual way to untangle them. Separate Crosstalk removes that dead end.

To recap, you can:

  • Recover usable dialogue from a moment where two speakers talked at the same time, without a rerecord.
  • Work with mono or stereo source, including phone and single-mic recordings.
  • Get separate Speaker 1, Speaker 2, and background tracks to mix independently.
  • Target just the overlapping section of a clip to keep credit cost down.
  • Process clips from 1 second up to 10 minutes.
  • Trigger it directly from the clip or the Properties panel, no extra setup.


Try It Out Now

This feature is available now in the latest Premiere (beta) build (26.5.0, Build 75). Give it a spin on real crosstalk moments from your own footage, whether that's a run-and-gun interview, a documentary scene, or anything shot on a single mic. Let us know how it feels and what you want to see next. Community feedback has shaped Separate Crosstalk, so your feedback directly shapes where this feature goes next!


Happy editing!

Note: We're still polishing some interactions, so expect to see improvements as we iterate.

 

    6 replies

    BeratGL
    Known Participant
    July 30, 2026

    Is there any plan to bring a similar feature for music? It would be incredibly useful to separate individual elements such as vocals, drums, bass, and other instruments directly within Premiere. That would make audio cleanup and remixing much more efficient. Is this something that's being considered for a future update?

    Adolfo H.Community ManagerAuthor
    Community Manager
    July 30, 2026

    Hi! this is in our radar :) I’ll keep you posted once we have updates. Thank you!

    FlyingFourFun
    Legend
    July 30, 2026

    When do you think some sample files and outputs will be available for review?  I'd like to see this, but dont have a track with a problem to try right now.

    Adolfo H.Community ManagerAuthor
    Community Manager
    July 30, 2026

    Hey! Please send me a DM and I can send you some files to try out, or we can jump on call so you can see it in action. Thank you!

    Stan Jones
    Community Expert
    Community Expert
    July 29, 2026

    @Adolfo H.,

     

    Wow. Just Wow. Some AI advances seem “ho-hum”; and then ones like this that are magic. Congrats to the team.

     

    I tested on Win 11 PR Beta 26.5.0.77.

     

    I didn’t have a real life crosstalk sample handy, so I created one and exported as a single 94 second mp4 with a stereo track with both speakers on both channels. There were 2 overlaps, one about 12 seconds and one about 8 seconds. 

     

    Credits used were in line with the one second per credit. The popup dialogue tells you what to expect, and gives the option to see your current balance. The estimate and actual were within one credit. The full 94 seconds (plus a few frames) used 95 credits, and the smaller ones 22 and 18.

     

    Once selected, you see uploading, processing, downloading. The entire processes are quick.

     

    As described, the result is 4 audio tracks, the original A1, which appears disabled, and the 2 tracks with one speaker each, and background. (I had no background sound, and this track appeared empty.)

     

    My quick conclusion was that, while I got good separation in the single 94 second process, the non-overlapping sections did not always get the correct speaker, and transcription would show up to 4 speakers. So I would always identify the overlap to avoid extra work isolating the separated overlaps later.

     

    My primary interest is in getting clean transcriptions per speaker. So to get the 2 overlap instances fully isolated, I created a sequence for each one. Trimmed the clip to the overlap 1 in sequence 1 and to overlap2 in sequence 2, and applied “separate crosstalk” to each one. 

     

    In the Project Panel, a bin is created for each crosstalk instance, and there is a separate .wav for each of the 3 tracks. I selected all 4 of the voice tracks created by my 2 sequences, the transcribed them as source clips. Each is available in its sequence.

     

    The major problem is one already at issue in regular, multitrack text-based editing: only the uppermost audio track shows its transcription. The workarounds are cumbersome. I don’t find an official feature request for this in my notes. Creating captions from overlapping transcriptions is an issue. Keeping transcript speaker identification when captions are created will be incrasingly important. But separating the crosstalk is an important step!

     

    Stan

     

     

     

     

    Adolfo H.Community ManagerAuthor
    Community Manager
    July 29, 2026

    Thank you so much for this feedback!!! We are working on a way to improve the non-crosstalk speaker embedding so they show up in the right clips. Our primary focus was to get the crosstalk to work. I’ll keep you updated once we have new updates on this regard.
    I’ll take notes on the rest of your comments and pass them to the team! Thank you for trying it out!