Setup guide
A voiceover recording setup that starts before the software
Most of what gets blamed on plugins is really a room problem or a gain problem. The honest voiceover chain runs in this order: fabric in the room, a preamp set with headroom, a small processing rack with no gate in it, monitoring you can trust, and a recorder pointed at the finished bus.
The room decides more than the rack
Every voiceover you admire was recorded close to a mic in a dead space. A hard-walled room reflects your voice back a few milliseconds late, and the mic captures voice and reflections as one signal. That boxy, roomy color is baked in: no EQ move separates the reflection from the word it is glued to, because they are the same frequencies arriving twice.
Which is why the cheapest upgrade in this entire guide is fabric. A duvet hung behind the mic, a closet full of hanging clothes, a couple of moving blankets on stands around your first reflection points: any of these beats any plugin on this page. We make audio software, and this is still the advice. Then work close, roughly four to eight inches from the mic and slightly off-axis. Every inch closer raises your voice against the room, which buys you more quiet than any noise tool.
Set the gain once, at the interface
Gain staging for voiceover happens on one knob: the preamp gain on your interface, before any software sees the signal. Perform your loudest line, the real one with the laugh in it, and turn the knob until those peaks land between -12 and -8 dBFS on the meter. The asymmetry is the whole logic: a take that is a little quiet is completely fixable, since patchd's free Gain node adds up to +24 dB cleanly if a quiet mic needs it. A take that touched 0 dBFS is not, because clipped samples are written flat at full scale and nothing restores what the converter never stored.
The voiceover rack: high-pass, EQ, gentle compression
The rack order is the signal order, and cleanup comes before anything that measures the signal, for the reasons the chain-order guide works through. The High-pass node goes first: rumble from traffic and desk bumps is barely audible as tone, but it is real energy, and it trips the compressor's detector if you leave it in. Roll it off around 80 to 120 Hz before anything downstream gets to react to it.
The EQ makes small moves: a gentle cut around 250 to 500 Hz if the take sounds boxy, a modest presence lift around 3 to 6 kHz for clarity, and not much else. The walkthrough in EQing your mic covers finding those spots by ear.
The Compressor's shipping defaults are already a narration setting: threshold -18 dB, ratio 3:1, 10 ms attack, 100 ms release, a soft 6 dB knee. At those numbers a peak that hits -6 dB comes out 8 dB quieter, and quiet words pass untouched, which is exactly the gentle evening-out a read wants. If it sounds like it is working, back the ratio off toward 2:1; the compressor guide goes deeper.
Where the gate went: nowhere, on purpose
A noise gate earns its keep live, on a stream or a call, where silence has to be sealed in real time because nobody gets a second take. Voiceover is the opposite situation. The silence between your takes costs nothing, because you will cut it in the edit anyway, so the gate's one job has already been done by your editor: you.
Meanwhile the gate's honest downside gets worse on narration. A gate judges loudness and nothing else, and spoken lines often start soft: a breathed h, a word that fades in. Those onsets sit below the threshold, the gate opens a beat late, and the front of the word gets clipped off. On a busy stream nobody notices; in a quiet read, it is a retake. If a steady hiss under the words bothers you, the tool that removes floor everywhere, including under speech, is noise suppression, not a gate, and the pauses still belong to the edit.
The limiter: a seatbelt at -1.0 dB
The last row in the rack is insurance. A limiter is a compressor whose ratio approaches infinity at a hard ceiling, and patchd's ships with that ceiling at -1.0 dB. Its job is to do nothing for the whole take and catch the one laugh you did not plan. One honest engineering note: patchd's limiter is deliberately zero-latency, with no lookahead. Studio mastering limiters buy perfect transient catching with 1 to 5 ms of lookahead delay; patchd accepts slightly less perfect catching so the node is safe in a live monitoring path. And if you ever see it working hard on an ordinary sentence, the problem is not the limiter: your gain staging drifted, so go back two sections.
Hearing yourself while you read
Most voice actors monitor themselves, and monitoring is only useful when it feels instant. In patchd, light the green pill under your mic strip's HW header to route the strip to your headphones. Latency through the bus is about 10.7 ms at a 512 buffer, and smaller ASIO buffers take it lower. The buffer itself is set in your interface's own control panel, not in patchd; the master clock readout at the top of the app shows the locked source, sample rate, and the live measured result, so you can verify what you are actually getting. If self-monitoring still feels off, see monitoring your mic with low latency.
Record the bus, not the raw mic
The last link in the chain is posture: patchd routes, and your recording app captures. Light one cyan pill under the mic strip's BUS header to send the processed voice to Bus 1, then open your recorder, a DAW or any app that can record from a microphone, and set its input to Patchd Bus 1. That single choice bakes the whole chain into the capture: the high-pass, the EQ, the compression, the -1.0 dB ceiling, all of it rides along, because to the recorder the bus is just a very well-behaved microphone.
Where patchd is today
patchd is built for exactly this kind of chain: one strip for your mic, free nodes for the cleanup, honest numbers on the clock, and buses your recorder captures like microphones. It is in development now. Join the waitlist and we will tell you the moment it is ready to install.