Parallel compression blends two versions of the same vocal: one dry or barely touched, one squashed hard, mixed back together on a bus. The squashed path pulls the quiet consonants, breaths and word-tails up to a usable level. The dry or lightly compressed path keeps the loud syllables sounding like themselves, transients intact. That is the opposite of what a single compressor on the insert does, which pulls the loud parts down and leaves the quiet parts exactly where they started. Both approaches raise average level. Only one of them adds density without also flattening the performance's dynamic shape.
What actually happens when a dry and a compressed signal are blended?
Route the vocal to a bus, print a heavily compressed version on that bus, and mix it back under the original at whatever level sounds right, often 6 to 10dB below the dry path to start. On the loud syllables, the compressed path is squashed so hard it sits well under the dry signal's full level. On the quiet syllables, breaths and word-tails, the compressed path has pulled everything up close to the same ceiling, so those moments read close to full loudness while the dry path has already fallen away. The blend is louder everywhere it used to be quiet and barely different everywhere it was already loud.
Why does this preserve transients that one compressor on the insert would flatten?
Because the transient-carrying path, the dry or lightly compressed one, never gets squashed. Parallel compression is a form of upward compression: it raises soft passages rather than lowering loud ones, and the human ear is far more sensitive to a loud sound suddenly getting quieter than to a soft one getting louder, which is a large part of why the trick works without announcing itself as compression. A single insert compressor set to do the same amount of gain reduction has no dry path to lean on. Every transient on that track passes through the same gain-reduction curve, loud syllable or soft one, and the attack and release settings shaping the quiet moments are also shaping the loud ones.
What does the version with five compressors actually add?
Sound On Sound's breakdown of Michael Brauer's vocal chain describes feeding one vocal to five separate compressors in parallel, post-fade, each a different model chosen for a contrasting tonal character rather than a specific gain-reduction number. Two run comparatively clean. Three are picked for their distortion character at high drive. All five outputs sum to one master bus. The point is not more gain reduction, it is that different compressor topologies color a signal differently as a side effect of compressing it, and blending several gives more tonal range than one compressor's single flavor however it is set. Michael Brauer built this approach mixing vocals for the Rolling Stones, Aerosmith and Coldplay, among others, which is why his vocal chains get cited by name rather than by a single setting.
What is the actual risk the title is asking about?
A compressed parallel path raises everything quiet, not only the parts worth raising. Room tone, breath noise, mouth clicks, the tail of a consonant, the hum of a preamp run at high gain all sit in the same low-level territory as the material the trick is meant to bring forward, and a bus compressed hard enough to be useful raises the unwanted material by the same amount. A vocal recorded in a treated space with a clean gain stage tolerates aggressive parallel compression well. The same setting on a vocal recorded in a live room, or one that needed 15dB more preamp gain than usual, brings the room up to the same relative loudness as the singer. At that point the vocal sounds louder and the recording sounds worse, and the fix is not a different compressor. It is a noise floor the technique should never have been asked to hide.
Where should the high-pass filter go?
On the source, ahead of the split, not on the compressed bus alone. Mike Senior's writeup for Sound On Sound notes that processing meant to affect both paths equally, like a high-pass to remove rumble and plosives, belongs on the source track ahead of the sends, so it reaches the dry path and every compressed path identically. Put it only on the compressed bus and the dry signal keeps its low-frequency rumble while the compressed signal loses it, which shows up as a tonal mismatch between quiet and loud moments that is hard to place until the routing gets checked.
How does digital latency change the setup compared to an analog console?
It requires matching processing stages on both paths, not just matching levels. On an analog desk, a dry signal and a compressed signal reach the summing bus at effectively the same instant. In a DAW, a compressor plugin adds latency, typically a fraction of a millisecond to a few milliseconds depending on the algorithm, so the compressed path arrives slightly later than an unprocessed dry path. Two versions of the same signal, summed with a small time offset between them, produce comb filtering rather than density. The fix is to give the dry path its own compression stage too, set to little or no gain reduction, so both paths run through an equal number of processing blocks and arrive aligned. This matters more on a vocal than on a drum bus, because comb filtering reads as a hollow, phasey quality on a single sustained tone in a way it does not on a transient-heavy source.
Frequently asked
What is parallel compression, and why is it also called New York compression? It blends an unprocessed or lightly compressed vocal with a heavily compressed copy of the same signal, raising the quiet parts without touching the peaks. New York studio engineers were known for relying on it.
How is that different from putting one compressor on the vocal insert? An insert compressor pulls the loud peaks down. Parallel compression pushes the quiet parts up while a dry or lightly compressed path keeps the peaks intact.
Does it require multiple compressors, the way Michael Brauer uses five? No. One heavily compressed bus blended under a dry signal is parallel compression. Five compressors, each for a different tonal character, is a more elaborate version of the same idea.
If a rough mix needs a second opinion before a session moves forward
Send the dry and compressed levels separately, with a note on what each bus is doing. Full credit list at /credits, background and contact at /about.