Remote podcast interviews have become the default production setup. Two people, different cities, each recording their own audio locally, then combining the tracks in post. The workflow is standard. The technical problems it creates are also standard -- and they compound in ways that are not obvious until you are hours into an edit wondering why the conversation sounds off.
Three problems come up in nearly every remote multitrack session: clock drift, mic bleed, and track alignment. They are related but distinct, and the fixes for each are different.
Track drift: the slow problem you do not notice until the end
Two computers recording audio independently do not run at exactly the same clock speed. Even devices that nominally operate at 44.1kHz sample rate have slight variations in their actual timing. Over a short recording, this is imperceptible. Over a 90-minute interview, those tiny timing differences accumulate.
A drift of 0.01% over 90 minutes translates to approximately 5.4 seconds of desynchronization between the two tracks. The conversation that was perfectly in sync at the start of the interview has the guest's responses arriving noticeably late by the end. A simple alignment sync at the beginning of the session does not fix this -- you need to re-sync the tracks as a continuous correction across the entire file length.
This is a problem that manual editing handles poorly. The traditional approach is to cut the longer track into sections and nudge each section to stay in sync. Doing this manually every few minutes for a 90-minute file takes a long time and introduces cut points that can create audible artifacts if the alignment is slightly off.
Correcting drift algorithmically requires measuring the divergence continuously and applying a time-stretch correction that varies along the length of the file. The goal is a smooth correction that accounts for the actual accumulated drift without creating the step-function behavior of manual section-nudging.
Mic bleed in remote sessions
Bleed in remote recording is different from the in-person bleed problem. In remote sessions, bleed happens when a participant is listening through speakers rather than headphones. The host speaks, the guest hears it through their speakers, and the guest's microphone picks up a faint echo of the host's voice. When the guest's track is processed, you occasionally hear the host's voice bleeding through.
This is why "always wear headphones during recording" is the first piece of advice every remote recording guide gives. Headphones route the audio to the guest's ears without it reaching the guest's microphone. Most experienced podcast guests know this. New guests often do not, and you end up with bleed that has to be managed in post.
The signal processing challenge with remote bleed is that the bled audio is spectrally similar to the target voice -- they are both human speech. Standard noise reduction, which works by modeling a noise profile of non-voice content, does not cleanly separate two voices sharing similar frequency content. The approach that works better is phase-based cancellation: because the bled signal arrives at the microphone slightly delayed relative to the original, there is a phase relationship you can use to reduce it.
This does not produce a perfect result. Heavy bleed from a loud speaker monitor cannot be fully removed without also affecting the target voice. The practical lesson is that preventing bleed during recording is worth far more than trying to fix it in post.
Track alignment: getting the sync right before anything else
Before you can work on bleed, drift, or any other processing, the tracks need to be aligned. For remote recordings, alignment typically starts with finding a common audio event at the beginning of the session -- a spoken countdown, a handclap, or a specific word that both microphones captured -- and using that as the sync point.
The challenge is that common audio events in remote sessions are never identical on both tracks. The host's voice arrives at the guest's microphone delayed by network latency, and the guest's microphone captures it at a different amplitude than the original. The guest's affirmation ("ready") arrives at the host's microphone under the same constraints. You are aligning peaks from two recordings that are similar but not identical.
A robust alignment algorithm finds the best correlation between the two tracks rather than looking for an exact match. Cross-correlation analysis over the first 30 seconds of the session finds the offset that produces the highest similarity between the two signals -- not a perfect match, but the offset where they are most in phase with each other. That becomes the initial alignment point, after which drift correction handles the rest of the file.
The combined problem in a typical remote session
In a typical 60-minute remote interview with headphone-wearing participants and reasonable network conditions, the problems look like this: initial alignment offset of 0.2 to 0.8 seconds (network and local recording latency), accumulated drift of 1 to 3 seconds by end of session, and minimal bleed because headphones were used.
In a session where the guest used speakers, add bleed management to that list. In a session with significant network instability, add correction for audio dropouts and reconstruction of missing segments -- a harder problem that is not always solvable cleanly.
Processing these problems in the right order matters. Align first, then correct drift, then handle bleed. Applying bleed cancellation before alignment produces results based on a misaligned signal, which makes the bleed model inaccurate.
What this means for your recording setup
The most effective thing you can do to reduce multitrack headaches is to record a double-ender: both participants record locally to high quality, both recordings are shared after the session, and you have the full-quality local audio to work with rather than a network-compressed stream. This is now the standard for serious podcast production and most guests will have heard of it.
Require headphones. Use a handclap or countdown at the start for easy alignment. For sessions that run long, a second sync point at the midpoint (a handclap during a natural break) reduces the amount of drift you have to correct algorithmically.
None of this eliminates the technical problems entirely. It makes them small enough that the tools can handle them without requiring extensive manual intervention.
