It is essentially IMPOSSIBLE to use Descript for transcriptions of live conversation podcasts because of this problem. It's incredibly frustrating.

Yes - in the transcript - when people on different tracks are speaking at the same time then show the transcript for each track side by side in vertical columns for the period that they are talking over each other. then when only one person is speaking return to the normal transcript

I'm primarily working with properly isolated multitrack dialogue recordings, and yet the transcription process seems to favor "oh" and "hmm" and "yeah" and even incoherent mumblings over people who are clearly speaking. Transcription ends up peppering a section of monologue with bits of other speakers — who, again, aren't really speaking so much as interjecting. As Ryan says, some way to flag a section as "crosstalk" would be excellent.

I'm surprised this isn't a more popular request. I edit a podcast with six people who record remotely, and cross-talk is my most significant time sink. Without a better way to handle this, the Descript composition view is little better than a single-track DAW. I can't save time by reading or scanning the text because I can't trust the text on the screen. I can't save time with the tools for removing filler words or unnecessary gaps because I can't trust it'll handle cross-talk. This tool could save me so much time (and money) if the composition view could be more than just a single-track editor.

This definitely needs a more elegant solution. The only way I fix it now is to edit the sequence manually to remove the cross talk. And it's not exactly the smoothest editor. :)

a similar product called Sonix can handle this with relative ease, so it confuses the hell out of me that Descript can't handle it.....considering it already uploads files individually. Someone on the design-train decided compositions needed to attempt to transcribe all audio together instead of separately and it shat the whole program.

We need a way to re-trigger a 2nd transcription for a problematic section after fixing the cross talk in a sequence.

+1 Even when I have individual speakers on separate tracks in a sequence, the speaker label assignment gets confused by cross-talk and tends to omit the transcript for that section entirely.

When this happens there are either no speaker labels at all, or incorrect speaker labels assigned to the tracks. The transcript indicates silence for several seconds, and any manual attempts to correct the transcription by adding the words being spoken usually makes the transcript worse by removing more sections of previously-detected transcript.

Two ways this could be handled: Let us mark instances of cross talk and deactivate auto transcript on those parts. Let us humans handle it. Or on the whole file, give us options to manually deactivate/force transcribing. It seems like the app keeps trying to retranscribe on its own whenever I make an edit. That's super annoying.