The more work I do, the more I’ve come to accept that bleed can actually be beneficial. 

…Sometimes. Like anything else, it depends on the context. Bleed in a live performance creates live mixing problems. Bleed in a live recording, however, can (sometimes) help the recording sound more natural.

I’ve done a not-insignificant amount of live mixing at this point in my life, and on a small-ish stage, bleed is a mix killer. Where I work, the stage is big enough to fit a band comfortably, but everyone is still close together. The drums are only a couple of feet behind the singer, the ceiling is relatively low, and the room is concrete and parallel surfaces. Pushing the vocals above the stage volume brings the cymbals and snare, etc., up with them. Sometimes, the bleed in the mic is as loud as or even louder than the vocals (soft singer, loud stage - or a singing drummer), so I don’t see gating as a viable option. I joke that every mic on the stage is a drum mic and do my best to ride the faders down if the backup vocals are not vocalling. I up the vocals in the PA understanding that it comes with a cost.

In a live context, in a smaller space, the mix doesn’t come from the PA alone – it happens in the room. What I do at FOH is primarily sound reinforcement. The sound that comes from the stage is the heart of the live mix, so the best possible mix that can exist in the room comes from a well-balanced stage. 

So then what happens when you take a multitracked recording of a live performance and mix it outside of the room? Outside of the real-time experience of the show, with no amps or drums pushing air in front of you? How do you use close mics to create a mix that sounds alive, that captures the energy of the moment, that offers clarity where it was lacking before, but that still sounds like it happened there?

Bleed can help. Usually. I’ll get to that in a minute.

I went through a phase years ago where I was obsessed with using iZotope RX to essentially strip the room out of the recordings that I was making at home. If the room doesn’t sound good, remove the room, right? Well, aside from introducing the undesirable artifacts of such processing, I soon realized that I was removing the context of the recording. Even if the room doesn’t “sound good,” the context is important. Recorded sound feels unnatural without context.

When I start a new mix from a live multitrack, the first thing I do is make sure everything is as time-aligned and phase-correlated as possible (I use Auto-Align 2 from Sound Radix - not sponsored, just find it to be an incredibly helpful tool). Then I start with the lead vocal and build the mix from there. My approach is to create a solid mix from the close mics and then bring the room mics up underneath to give a sense of the space, but not so much that the mix gets washed out. It doesn’t take much, and I end up having to clip gain/automate applause up to make it sound more like it did in real life, ie not so quiet and far away. Maybe if the space sounded better I could lean more heavily on the room mics when constructing a mix from the recording, but the room is a concrete and plaster shoe box, so for the sake of an intelligible mix, I prefer to keep its presence minimal. Preserving the impression of the room is crucial, though.

I will tell you that because I’ve identified vocal mic bleed as my nemesis while mixing live, my first inclination was to run software on the vocal recording to extract the vocal from the bleed. I’ve tried iZotope and I’ve tried SpectraLayers (heavily endorsed by the man Andrew Scheps) for vocal isolation. Although they both introduce artifacts, I found SpectraLayers to be a little more agreeable to my ears. Initially, I thought that the first rock mix I handled this way turned out really well, but when I listened back to it later, I felt like the vocals were coming from a void. Even in a dense mix, it felt odd to me. Something was missing.

I was wondering how KEXP and Audiotree do it, and in both cases, I found interviews/forums/etc. where those engineers talk about embracing the bleed as part of the sound. Coincidentally, I’ve recently watched a lot of videos of Philip Weinrobe explaining his recording and mixing philosophy, and he’s a huge advocate for embracing bleed (which he prefers to call “spill”). So the bleed stays.

Since then, I’ve experimented with what I consider parallel isolation, where I’ll use SpectraLayers to “unmix” the voice from the bleed, then I’ll heavily compress the isolated vocal and blend it back up under the original vocal recording as support (still have to be careful about artifacts when there’s applause or talking between songs). This is a technique that I’ll continue to refine, but it’s worked well for me so far in a rock setting – it gives the vocal more authority in a big mix, but it allows the recorded bleed from the performance to stay as glue. Preserving the bleed and not being afraid to push instruments up around the vocal (another lesson I’ve learned) means no more void.

The most challenging recordings I’ve faced in terms of bleed are the ones that should ostensibly be the easiest to mix: solo singer-songwriters with DI’d acoustic guitars. In a controlled studio environment, with thoughtful mic choice and placement, bleed can and does work to shape the character of the recording. In a live recording context, bleed is lethal. Boosting presence frequencies on the vocals also boosts unflattering clang on the guitar, which is especially problematic when the performer is an enthusiastic strummer. The result is a high-mid smear that distracts from the vocal. And you might think that a DI would be a clean guitar signal (we’ll ignore the general timbre of an acoustic DI recording), but consider that an acoustic guitar projects sound because it’s a finely tuned resonating body, and then consider that it’s held close to the singer’s chest… Gaining up the guitar in quiet sections means gaining up any vocal resonance that might be present in the signal, which then sounds like vocal feedback layered under the vocal track. It’s so challenging because each element is glaringly exposed in the mix, and with nowhere to hide, artifacts from isolation processing are way too obvious for anything but quick spot fixes. I feel like I’m wrestling the whole time. (Side rant: I know, I could use a microphone on the guitar in addition to the DI, but these artists often perform standing up, and they move the guitar around a lot, so with the relative position constantly changing, I imagine the microphone signal would just add to the headache at mixdown. I won’t ask an artist to tone down their performance for the sake of the recording because the absolute most important thing is that the artist and their fans have a positive experience sharing the space together in real time. The recording is auxiliary to that. It’s a show, not a recording session.)

Whew, okay, rant over. Now we’ve reached the part where I talk about the multitrack that inspired me to actually write another blog post. It’s a situation that’s requiring me to use what I’ve learned so far about bleed and spectral processing to address a specific set of challenges. For context, it’s a rock band with drums, bass, electric guitar, keys, keys bgv mic that is also auxiliary percussion, and bass bgv mic that is also flute. The lead singer used effects pedals on his vocals, and that’s what was recorded. The background vocalists wanted a huge amount of reverb and delay, but the recorded signal I have is dry. (Note: I did not employ the parallel isolation technique on the lead vocal for this mix because the recorded signal was workable.)

Challenge 1: Background vocal reverb and drum bleed

I’m adding reverb and delay to the recorded background vocals like I did during the show, which means that all of the bleed is positively swimming. It wasn’t obvious during the live performance because there was so much drum sound coming from the stage, but on the recording, the added reverb is overpowering. I don’t want to totally strip the bleed for the reasons discussed above, so I could automate the effects sends to open only when singing is happening, but because the bleed is about as loud as the intended source, it’s very, very obvious when those sends are open. I don’t want the snare to suddenly have a big reverb tail when the background vocals come in.

Solution: “Unmix” the bgv recordings using SpectraLayers, import ALL layers back into the session WITHOUT DISCARDING ANYTHING, and only send the separated vocals and flute (and even the aux percussion) to the effects. Bleed preserved, but only the intended signal is sent to the effects. Keeping all of the layers that SpectraLayers identified means that as long as the relative volume is the same across the layers, artifacts should be minimal.

Challenge 2: The (un)realism of DI keys

This one goes back to recorded sound and context. It’s a more subtle issue than the drum reverb, but it’s one that I want to address to help the recording feel natural. The keyboard player used an amp onstage, but the tracks that I have (Nord and Juno) were taken as direct outs from the amp. They have no inherent context, no air around them. The room mics provide some context, but as previously stated, I blend those in conservatively. The keys need a little more than that to help remove them from their vacuum and gel into the mix. Think about a live recording as a painting: each instrument/voice/source is its own color, and during the live performance, the colors blend together at the edges as they combine in the air. The paint is wet. Later, when you bring the fader up on a DI signal in the recording, it’s like making a brushstroke on top of the painting after the other colors have dried. There’s a distinct boundary to the sound that prevents it from merging with the rest of the composition.

Solution: Preserve the bgv mic bleed! Is it the most flattering depiction of the keys? No. But it provides an angle that we don’t have from the DI alone. Air, not just electrons.

What’s the big takeaway from all of this? Don’t be so afraid of bleed. It seems counterintuitive because we’re generally fixated on clean recordings, but bleed really can help in certain scenarios. You can find creative ways to use it to your advantage. 

(PS: If you have a tried-and-true solution for my DI singer-songwriter woes that does not include secretly re-recording the guitar myself at home, pleeease share. I have not done that and I don’t intend to, but that’s about the only thing I can think of.)