Setup guide
Remote podcast guest audio: a clean lane for a mangled voice
A remote guest reaches you through a call app, which means their voice arrives compressed, gain-ridden, and occasionally missing a word. You cannot un-mangle it after the fact. What you can do is stop making it worse: ask for four things at their end, give the call its own lane at yours, and record that lane as its own track.
What the call app actually delivers
Every call app is built for conversation, not recording, and the engineering trade-offs show. A lossy voice codec throws away parts of the signal to fit the connection. The app's own noise suppression chews on the room, and sometimes on the ends of words. Automatic gain control rides the level up and down so everyone sits at roughly the same loudness, flattening the difference between a whisper and a laugh. And when the network hiccups, packets simply never arrive: a syllable, a word, sometimes a whole phrase is gone, and nothing on your side ever received it.
None of that is recoverable downstream. No EQ can re-synthesize what a codec discarded, and no plugin refills a dropout. So the honest playbook has three moves, in order of leverage: raise the quality at the guest's end before the codec touches it, keep the call on its own lane at your end so the damage stays contained on one editable track, and when the episode really matters, do not record the call at all: have the guest record themselves. Each move is a section below.
Four things to ask of your guest
The biggest wins happen on hardware you do not control, so send the checklist before the recording day. None of it requires the guest to buy anything beyond, at most, a modest microphone.
- Headphones, any headphones. The moment your voice plays out of their speakers, their call app has to run echo cancellation to claw it back out of their mic, and that processing audibly degrades everything they say. Earbuds end the problem entirely.
- A mic near their mouth. A headset mic or a cheap USB mic a hand-span from their lips beats an expensive laptop array across the desk, because distance is room reflection, and room reflection is what the call app's suppression mangles hardest.
- A quiet, soft room. Closed window, fan off, curtains and cushions over bare walls if there is a choice of rooms. Quieter input means the app's noise suppression has less to do, and suppression that does less does less harm.
- Wired over wireless, twice. Ethernet instead of Wi-Fi where possible, because dropouts are network events. And a wired headset instead of Bluetooth, which stacks its own lossy hop on top of the call's codec.
Your side: the call already has its own strip
In patchd, there is nothing to set up to receive the call. The mixer keeps a fixed Communications strip, and audio from chat apps lands on it automatically, whether the call runs in Discord, Zoom, or Teams. No device picker, no drivers to point at each other: the guest arrives on their own strip with their own fader, meter, and effects rack.
The routing takes two pills. Light Playback so the call reaches your headphones, then light Bus 3 to give the guest a lane of their own. A bus is a virtual output that other apps see as a microphone, so lighting that pill is what makes the call separately recordable. Keep Bus 3 dark on every other strip: a bus carries exactly the strips routed into it, so the lane stays pure only while the Communications strip is the only one feeding it.
One honest limit: this is a lane per call, not per caller. The call app hands your PC a single premixed stream, so two remote guests in the same call arrive on the same strip and no mixer can pull them apart afterward. If you need two remote voices on two tracks, that is a job for the double-ender below.
Record the lane as its own track
patchd routes; your recording app records. Every patchd bus shows up in Windows as a microphone, so open your recorder or DAW, create one track per lane, and point each track's input at the matching bus: your own mic's bus on one track, the guest lane on another. The in-room half of this layout, one strip and one bus per physical mic, is the multi-mic podcast guide; the guest lane simply joins it as one more track.
Even with a codec-flattened voice, the separate track earns its keep. The damage is contained: your voice stays clean on its own lane no matter what the network did to theirs. Level fixes are per voice, so pulling the guest up never drags your track with it. And your edits stay one-sided: a chair squeak on your lane cuts without touching their answer. The rack on the Communications strip can take the edge off before recording, a gentle EQ for boxiness, but be honest with yourself about the ceiling: processing shapes what arrived, and cannot restore what the codec discarded.
The double-ender: the honest upgrade
When the episode matters, stop recording the call and record the person. In a double-ender, the guest hits record on their own machine, any recorder they have, even a phone's voice memo app, while you both talk over the call as usual. Their file captures the voice at their end, before the codec, the suppression, and the network ever touch it. After the call, they send you the file and it replaces the call track in your edit.
Your Bus 3 recording does not become useless; it becomes two other things. It is the safety net: if the guest's recording fails, was never started, or arrives clipped, you still have a complete episode. And it is the sync reference: both recordings contain the same performance, so you slide the local file against the call track until the words line up, then mute the call track and keep it in the project. Be clear-eyed about the division of labor here: the double-ender happens in the guest's recorder, not in patchd. What patchd contributes is the clean, isolated call track that makes the sync trivial and the fallback real.
Where patchd is today
patchd is built for exactly this shape of problem: a fixed strip that catches the call automatically, color-coded buses that turn any source into its own recordable lane, and a rack per strip for the cleanup that is honestly possible. It is in development now. Join the waitlist and we will tell you the moment it is ready to install.