The success of Kesha’s ‘Praying’ justified the lengths that Jon Castelli went to to get the perfect mix — eventually!
“There are no rules in mixing, because it’s not about things sounding good or correct. It’s about how things feel. As a mixer, I always try to go one step further in getting it to feel the best possible. If mix version 4 gave you goosebumps, I want version 20 to give you even more goosebumps. Going that extra mile is what separates a great mix from a mediocre mix. Many mixers are willing to put up with something mediocre because they don’t know what to do when they’re faced with several different opinions, from the artist, the producer, A&R person, management, and so on. But what I learned from someone like Tony is that there’s always a way to solve that problem. The solution is to do whatever it takes to make it feel better than anyone could have imagined. That attitude is more important than the techniques, or what plug-ins are used.”
Thus speaks Jon Castelli, and the Tony he refers to is one of the world’s leading mixers: Tony Maserati. Originally from New York, Castelli moved to Los Angeles seven years ago, where he was signed to Maserati’s Mirrorball Entertainment as a mixer, songwriter and producer. With Maserati as his mentor, and often working with him on a joint mix basis, Castelli clocked up mix credits ranging from Lady Gaga to Ariana Grande. Two years ago Castelli went fully independent, and he now operates from his own studio, called The Gift Shop, located above a boutique in the Downtown Los Angeles Arts District, and containing a rather unusual hybrid of digital and analogue gear.
In that quote, moreover, Castelli is not only describing his general mix attitude, but referring to a specific situation in which he didn’t have to go so much the extra mile as the extra marathon. The job in question is his mix of the lead single of Kesha’s third album Rainbow, ‘Praying’. An absence of five years left many people wondering whether the singer would ever make an album again, but Rainbow turned out to be a triumphant return, widely lauded by the critics and reaching the number-one spot in the US.
The fact that Castelli mixed the track at all also defied expectations, given that half or more of all charting pop singles are currently mixed by just two top mixers: Manny Marroquin and Serban Ghenea. Record companies tend to contract one or both of them to mix a pop album’s main singles, and give the job of mixing the other album tracks to slightly less well-known and presumably cheaper engineers. Because of where Kesha’s career has been over the last five years, her record company, Kemosabe/RCA, pulled out all the stops to make the new album a success, and Marroquin’s safe pair of mix hands were indeed on seven of the album’s 14 tracks. Others were mixed by reputable names like Shawn Everett, David Boucher and Chuck Ainley. In this company, Castelli is the new kid on the block, yet he mixed four songs including the lead single.
This came about through his connection with the main writer and producer of ‘Praying’, Ryan Lewis, of Macklemore & Ryan Lewis fame, which goes back to Castelli’s mix of the second Macklemore & Lewis album, This Unruly Mess I’ve Made (2016). It was a career breakthrough for Castelli, which allowed him to spread out on his own and create The Gift Shop. And so, one day in the middle of February 2017, Castelli was sitting in his studio when the phone rang. Ryan Lewis was on the other side, with an urgent request...
“Ryan said: ‘Can you please get on a plane in Burbank in three hours? I need to finish this Kesha song, and it isn’t quite right. It’s at a place where it’s close, but I’m not sure where to take it from here. Can you help me finish it?’ I was aware that Ryan had written most of ‘Praying’ a year before it was recorded, with Kesha in mind, but I had never heard the song before. So I flew over to Seattle and went to Ryan’s studio, which is in his house. I was sitting on the chair behind him when he played the session to me, twice, without giving me any notes or explanations. He tends to work in Ableton, but the session was by now stemmed out into Pro Tools. He asked me what I thought, and I replied that in my opinion the arrangement and the vocal treatment and comp needed work.
“In fact, my first thought after hearing Ryan’s rough was, ‘This does not sound intimate enough. How do we dial this back?’, and so the first thing I did was mute the reverb on Kesha’s vocal in the first verse, first chorus and second verse. The demo had a lot of reverb on the vocals, turning it into a kind of Whitney Houston-style ballad, but for me that sounded a little dated. Ryan is more used to hip-hop vocals in any case, and on recent pop songs by, for example, Lorde and Taylor Swift, there’s no reverb on the lead vocal, or if there is, it’s hardly noticeable. So, muting the reverb made it sound more modern, but my other thought was that the vocals in the first half of the song in particular needed to sound more intimate. When we later sent the mix to Kesha, she was like, ‘Oh, it sounds like it’s in my bedroom,’ and this worked really well for these parts of the song.
“After muting the reverb I also did a new vocal comp the evening I arrived, and the next day I spent a lot of time on the arrangement, muting parts and moving them around. Ryan and I went back and forth in discussing what was better and what was not, taking out quite a few things so that when they come in halfway through the song they have more impact, and also creating more dynamics. For that process I had my producer hat on, and any processing I did was more broad strokes. Ryan obviously knows what he’s doing, so the source material sounded great, but in a situation where I don’t think the sounds are good enough to decide whether they are worth using I might first focus on cleaning them up and get deep into using plug-ins. But in the case of ‘Prayer’ I worked on arrangements first, on the second day I was there, and then on the third day I did a rough mix. Part of that involved taking off some of the limiting, again because we wanted more dynamics in the song.”
In the screenshots of Castelli’s in-the-box mix from Lewis’s place in Seattle, a mix session referred to as v4 and dated February 16, 2017, there’s ample evidence of a radical rearrangement process. Several piano, strings and choir parts are greyed out, particularly in the first half of the song. Combined with Castelli’s dramatically different vocal treatment for Kesha’s voice, these changes steered the ballad-like song into different territory, and gave it a very different feel. Castelli’s interventions in the song, which also had writing input from Kesha herself and songwriter Ben Abraham and composer/arranger Andrew Joslyn, created a starker contrast between the verses and pre-choruses, which became more introverted and intimate, and the choruses and bridge, which are larger and more expressive.
Castelli and Lewis sent the mix off to Kesha and her management, and the response was immediate and powerful. “We wanted to know whether Kesha liked the arrangement and production like this,” recalled Castelli, “and she sent us an email back, saying that she was extremely excited about the new version. She called it ‘stunning’. Her management and record company also really liked it.”
That, in most situations, would have been that. However, perhaps unwisely, Castelli’s commitment to excellence led him to say to Lewis, “I’m going redo the rough mix at my own studio, and send it through my analogue gear. So I did, and I liked it better, Ryan liked it better, and Kesha liked it better.”
That, once again, would have been that. However, as so often happens when several people are involved, just when everyone thinks that all’s said and done, there’s always one person who will put their hand up and say, “yes, but”, usually causing groans in all around them. Castelli: “Kesha’s manager came back to us, and said, ‘I really like the new mix you did, but you know what, there’s something about the rough mix you did in Seattle that gave me goosebumps. I don’t know what it is, but please see what you can do about that.’
“So we talked about it. Ryan called the rough mix the ‘bedroom mix’, because it’s like you’re next to her playing piano in her bedroom, and the choir and strings are more in the background, and he called my new mix the ‘platinum mix’, because it sounds lush and big and like it was going to be on the radio and go top 20 and sell a lot of records. But we recognised that there was something about the bedroom mix, where even if the vocal wasn’t as hi-fi or as clear as in the platinum mix, it has a rawness that people respond to. The goosebumps came from that rawness, which my second version had lost to some degree.”
Clearly, Lewis and particularly Castelli, with their predilection for excellence and feel, didn’t take Kesha’s manager’s opinion as a reason to groan, but instead as an invitation to “make it better than anyone involved could have imagined”. Castelli had conducted his semi-analogue ‘platinum mix’ at the end of March and Kesha’s manager’s comments came in soon afterwards. What followed was a period of nearly six weeks, off and on, during which Castelli tried to incorporate Kesha’s manager’s comments, and come up with a final mix that was even better. This involved another 19 mixes, none of which delivered the required result.
“After these 19 mixes Ryan and I Facetimed and discussed what we could do. We ended up coming up with the idea of using both the rough mix and my analogue mix and editing them together. So then the challenge became to see which mix felt better for which sections. That became a pretty intense process of A/B-ing, in which my assistant Ingmar Carlson was a star. He created a master session of stems from the in-the-box and the analogue mixes, and I went through and muted and unmuted these stems in different sections. We were constantly comparing and adjusting things — and yes, there’s the risk that you’re chasing your own tail and don’t see the wood for the trees any more when you go into that amount of detail, but we simply wanted to take the mix one step further.
“Ryan and I are both writers and producers, and we both understand engineering, so we were looking at the song and the different mix options as a whole, through that prism. We ended up with the first verse and pre-chorus and second verse from the rough mix, where the lyrics and the feel dictated that it needed to sound raw and like Kesha was whispering in your ear, and where if you make it sound too good, you lose some of that intimacy and honesty. We took the second and third choruses and bridge from the analogue mix, which sounds gigantic and gorgeous, which is what the chorus needed to be.
“Throughout the entire mix process I was very focused on the lyrics. It was crucial that the sonics reflected their intimate and personal nature, and the shifts between verses and choruses, which went from very bedroom in the verses to very gospel in the choruses. We were trying to create something unique and different, so there also was no reference. I did not pull up any other tracks to listen to and compare. We just had to be what it is, with an intimate verse production and the choruses being bombastic and anthemic, with a choir stomping on the pews in the church.”
The rough mix of ‘Praying’ that was done in the box at Lewis’s place in Seattle is, by modern standards, fairly moderate in size, with 70-odd tracks, and exceptionally well-organised, with all tracks properly named and grouped and colour coded according to instrument groups, and all clips neatly topped and tailed. Castelli’s rough mix print and master track are at the top of the mix session, followed by nine drum tracks, 17 piano tracks, four strings tracks, nine keys tracks (including two solo violin tracks), 12 choir tracks, three lead vocal tracks (in pink), with a ‘Voc Bus A’ track and two aux effect tracks above them, and 12 harmony vocal tracks. The entire session is densely populated with plug-ins, totalling several hundred instances.
Immediately below the drums in the rough mix session is the main piano track, called ‘PIANO.declick’. This piano appears in the first and second verse, pre-chorus and first chorus, though Castelli pulled the first chorus section to another track for a slightly different treatment. Both tracks have the same four plug-ins on the inserts, which are the UAD API 550a EQ, UAD Fairchild 670 compressor, FabFilter Pro-Q2 EQ and UAD Brainworx Millenia NSEQ. Castelli explains: “The API was an EQ Ryan had added, which I left, although I may have tweaked it a little bit. The Fairchild which follows after that has a very specific effect on the decay of sounds. The way it releases in the first Time Constant mode sounds super-controlled. I like to use the Fairchild on reverbs and on instruments with a long sustain, like pianos. It interacts dynamically with long tails and sounds really great on pianos. The reason the Q2 follows the Fairchild is that I had already taken out some harsh low mids with the API before it, but the Fairchild put these back a little bit because of the tube harmonic processing. It sounds more musical, but I nevertheless dipped 250Hz with the Q2 to compensate.
“I also used the Q2 to implement a high-pass filter and roll-off above 12kHz. I don’t believe that you need anything on the piano above 12k, because that is where other stuff is going to sit, like a vocals. I do that to almost every instrument, I am pretty geeky with low-passes and high-passes. I think people forget about that and just stack up sounds at the top end, and then they sacrifice the loudness because there is unwanted energy at the top. Finally, the Millennia is doing some very minor shelf things, boosting at 56Hz and 16k. This may seem strange after I’ve just rolled off above 10k with the Q2, but it’s based on that Pultec filter/boost technique where you are getting a resonant peak. The filtering on the Q2 changes the way the boost works on the Millennia.”
Below the main intro piano track are a set of Yamaha and a set of Steinway piano tracks, and below them the strings. Each of these three musical elements is sent to its own submix group track, with similar plug-ins to the main piano. In the case of the Yamaha submix track, these are augmented by the UAD Ampex ATR102 tape emulation plug-in, a FabFilter Pro-MB and a send to a Waves Doubler, which, says Castelli, “I like to put on the keys most of the time to add some width and some detuning. It’s either that or the TAL Chorus LX.” The Steinway sub track adds the UAD SPL Transient Designer, presumably for more attack, but omits the Doubler. The ‘Strings’ group track regains the Doubler and has a couple of other variations, most importantly the use of the FabFilter Saturn, for more grit.
‘Texas’ is the group track for the choir, so-called because it was recorded in Texas, and it contains an impressively long signal chain including a whopping five EQs. Castelli: “This is a crazy chain, and really what it was about was to shape the sound, so it did not sound like the source material. We wanted the choir to sound like a Kanye record, ie. scooped in the upper mid-range, with a big 3k dip, while keeping air on top. The goal also was to add energy to the vocals and some dirt, which is why the Saturn is there, to grunge it up. The EQ3 and Massive Passive were there from Ryan, so I left them, and the Fairchild and Pro-MB glued the voices together.”
The lead vocal section of the session consists of three main vocal tracks, all called ‘JONTAKES’, each for a different section of the song, which all go to the ‘Voc Bus A’ group track above them. The three vocal tracks all have the same signal chain, and Castelli is at pains to point out the significance of their ordering: Antares Auto-Tune going into the FabFilter Pro-DS de-esser and Q2 equaliser, Avid’s BF76 and the UAD LA2A compressors, and the FabFilter Pro-MB. The insert signal chain on the ‘Voc A Bus’ is FabFilter Q2 and Pro-MB, UAD 1176LN, Pro-DS, Q2 and UAD Millenia NSEQ-2, and there are sends to a reverb and delay.
Castelli: “The ‘JONTAKES’ are duplicates of my vocal comp, and I then split that in three different sections, because her voice has a different character in each. She goes from bedroom-like to theatrical to having a diva moment at the end of the bridge, and the biggest challenge in this whole mix was making her sound like one person. The vocals had all been recorded on the same day using the same mic, but it needed quite a bit of work to get them to sit right in the track. The plug-ins on each of the three tracks are the same, but with slightly different settings. Each LA2A is doing something different to balance out the levels and control the peaks.
“On the ‘Voc Bus A’ track there’s surgical EQ from the Q2s, notching at 1kHz and at 4kHz to get rid of some harsh frequencies in her voice, so I can get it to sound as loud as possible. One Q2 is before and the other is after the 1176, to compensate for what the compressor does. I also really like the top and bottom boost from the Millenia, which is a beautiful-sounding emulation of a warm tube EQ. I love the sound of tubes in general, so often use tube emulation plug-ins, and this is one of them. I have another de-esser on the vocal bus group track, which is kind of a Rich Costey technique, where you’re boosting a lot of 10kHz and up, and then de-ess after that. It gives it a kind of pop crunch.
“I muted the sends to the reverb and delay aux tracks in the verses. I put on the main reverb plug-in, which is the UAD AKG BX20, after which I have the FabFilter Pro-MB and the Fairchild 670. The Pro-MB controls the dense and harsh mid-range of the vocal verb, and the 670 once again is there because of the way it controls the reverb tail. The delay is just a standard Avid Mod Delay III, with a quarter-note delay. Below the lead vocals are 11 harmony vocal tracks, each with a Trim and Auto-Tune and they all go to a sub track, which has again the Q2 and the 670, but also the Saturn for some saturation, as well as sends to the same Reverb and Doubler aux tracks as the lead vocal.”
Six weeks after Castelli finished his rough mix in Seattle, he set about trying to improve this mix, working in his own studio, The Gift Shop. One reason for doing this was that he wanted to send the mix through his analogue chains, consisting of 14 Hazelrigg Industries VLC-1 units, which are vacuum tube channel strips, designed and licensed by DW Fearn. In addition to this, Castelli also has the DW Fearn VT-7 tube compressor on his master bus, as well as the Rascal Audio Two-V preamp and the Kush Audio UBK Fatso compressor.
A-D and D-A conversion at The Gift Shop happens courtesy of Black Lion Audio Sparrow MkII’s and a Burl B2, while monitoring at the facility is provided by PMC IB2s. Because the studio also has a tracking space, there are several mics, particularly ribbons by AEA and BeyerDynamic, and an AKC C12 tube mic.
“The reason I have the tube gear is because it creates harmonic content that I don’t believe exists in the digital realm,” asserts Castelli. “The tubes also give me a wall to hit without clipping, and a big fat bottom end, and in a world that likes bright mixes, a little more headroom to give me more top end that’s warm and not brittle. They allow me to make different mix decisions and get a sound that’s more musical. I built my own, custom console using 12 of my VLC-1 units, which are based on a vintage RCA preamp tube circuit, with a Pultec-style EQ on every channel. Using the VLC-1 units does not change my mix process. I have everything calibrated to where I like it, and I kind of leave it. It’s more of a summing thing than a desk. My 12 VLC-1 units in the console all come up as inserts in Pro Tools. I used one of them as an insert on the lead vocal. I had my two other VLC-1 units on the mix bus for this song.
“I’d say that 80 percent of my mixes go through my tube gear, but sometimes it does not work. If someone says, ‘It sounds too tube-like,’ the first thing I do is take out the tube gear. I’m not precious about my analogue gear anymore, despite spending a lot of money on it! Moreover, since Pro Tools went 32-bit float I think it sounds great in the box. I don’t have a problem with creating depth and separation and clarity in the box. I don’t necessarily think that the tube gear gives me an advantage, it’s more a matter of personal taste.”
With 33 tracks, Castelli’s ‘platinum’ mix session for ‘Praying’ is, to some degree, a distilled version of his rough mix session. It’s called ‘InProg16’, meaning mix-in-progress number 16, and at the top are three mix print tracks, a mix bus track and a ‘Limiter’ track with the FabFilter Pro-L. Below that six group aux tracks, called Bass, Keys, Synth, Strings, Vocal and Drum. Below that there are stems of the toms, kick reverb and 808. Further down are four piano stems, a Mellotron stem, a strings stem, a synth stem, another string stem, a backing vocal group track, a ‘VoxParallel’ bus, and the stem of the lead vocal.
At the bottom of the session are seven aux effects tracks, mostly used for the vocals, with the same Reverb (AKG) and Delay (ModDelay) tracks as from the rough mix session, followed by aux tracks with the UAD EMT 140, the UAD EMT 250, TAL Chorus LX, Waves Doubler and UAD Culture Vulture. Leaving the additional aux tracks aside, the tracks in this session contain far fewer plug-ins than the rough mix session, illustrating that Castelli was using the rough mix as a starting point, and was tweaking it, rather than reinventing it.
Castelli: “The analogue mix session is based on the rough mix, with tracks that have already been processed and committed. In many cases I tweaked things very subtly, A/B-ing tracks with the rough mix to see if I could make it sound better. So, for example, I had a Doubler and a Culture Vulture on the sends on the Yamaha piano subgroup, to give the piano some subtle width and saturation. The main drum bus has the Transient Master, and the Q2, dipping around 150Hz, and the UAD SSL E-channel. In the group tracks, the ‘Keys Group’ has the [iZotope] Ozone 7 Imager, to widen the keys, but only in the mid-range and high frequencies. I’m narrowing the lower frequencies to make sure they are as mono as possible. It’s the same with the ‘Strings Group’.
“The plug-ins on the inserts of the main Lead Vocal track remain from the Voc Bus A track from the rough mix session, and have the same settings. The settings on the BX20 on the Reverb aux also are the same, but I’m adding the EMT 140 and EMT 250 reverbs on the vocals. The lead and backing vocals then go to the ‘Vocal Group’, on which I have the Pro-MB and the iZotope Ozone 7 Vintage EQ. Yes, that means that if you add it all up, the lead vocals have 14 plug-ins on the inserts — six on each of the original three tracks, six on the ‘Voc Bus A’, plus another two plug-ins on the ‘Vocal Group’ track. Add the eight plug-ins I used on the six aux tracks that the ‘Lead Vocal’ stem is sent to and you could say that I had 22 plug-ins on the lead vocals!”
Returning to the top of the session, Castelli notes: “The blue mix print track right at the top is my rough mix done at Ryan’s place, so I could A/B it and can make sure that I keep the integrity of the balances and panning and initial vibe. It was like: ‘Make it better than this, or use this!’ In cases where I mix a track for which I haven’t been involved in the production, I would have the rough mix that was sent to me at the top, so I can A/B what I’m doing against that. It’s what I have to beat.
“The mix bus on this mix has the UAD Shadow Hills Compressor, the FabFilter Saturn, the iZotope Ozone 7 Maximizer, the UAD Millenia EQ and the UAD Chandler Curve Bender. The Saturn is bypassed, but it actually still affects the sound. This is a secret! I use over-sampling on it with the HQ button, and it really helps lift the sound. The print of my mix then went through the ‘Limiter’ track with the Pro-L for extra volume. This is for reference for my clients, but I also tend to send the version that has gone through the Pro-L to the mastering engineer.”
Finally, as described earlier, after 19 attempts by Castelli to achieve the desired blend between bedroom intimacy and luxurious-sounding expressiveness, Lewis and he decided to splice the rough and the analogue mix together. Part of Castelli’s process in reaching 19 different versions of his ‘analogue’ mix was that he had been trying different kick sounds in the session, and he added “the big giant concert bass-drum stomp that is in the second and third chorus”.
In the separate splice session, some of these different mix versions are shown in dark blue at the top. In the end, after endless A/B-ing and discussions via email, the decision was made to splice the rough mix together with Castelli’s ‘InProg16’ mix, as can be seen in the bottom track: the rough mix is in pink and the ‘InProg16’ mix in green. This edit was then printed and consolidated immediately above, entirely in pink, and called ‘InProg20’. It was the conclusion of a very long and very laborious process, but Castelli maintains that he would not have it any other way.
“I loved the process that we went through to get this, as hard and as long as it took. This is Ryan’s first released song outside of his work with Macklemore, and I worked really hard to get this to be exactly the way he wanted it. We managed, and I recall that people liked it so much that there were discussions about whether this song should be the lead single or not. When it was indeed decided that ‘Praying’ would be the lead single, Kesha’s A&R offered me another seven songs to mix, three of which also ended up on the album. Two more of the two tracks I mixed will come out on a deluxe version of the album.”
All of which seems ample justification for Castelli’s desire to go the extra mile, or marathon.
A striking feature of Jon Castelli’s original mix is that FabFilter and UAD plug-ins seem to account for almost 90 percent of the processing. This, it seems, reflects Castelli’s general preference. “I also use plug-ins by iZotope, Softube, SoundToys and a few by Waves, but it’s true, UAD and Fabfilter plug-ins are my go-to plug-ins. I have the UAD Thunderbolt Octo Satellite with every plug-in by UAD. In general I don’t want to have to think about endless plug-in options, and I don’t have a lot of gear in my studio either. I just use what gets the job done. When it comes to weird effects and modulations and things like that I’m down to geek out and try any plug-in, but for basic EQ and compression I tend to stick with what I know.
“UAD is the best at modelling classic gear that adds character. FabFilter is the most forward-thinking plug-in company, so they are in many ways the opposite to UAD. You can modulate things with FabFilter plug-ins in ways that you can’t do with any other plug-ins. The Saturn saturation and distortion plug-in, for example, is super-geeky, with tons of unusual options and sounds. I also love the FabFilter interfaces, which often have literally nothing on the screen. You end up thinking of processing more like a producer than an engineer. And the presets are great. My use of many UAD and FabFilter plug-ins is like having a combination of analogue and digital in the box.
“I use the Pro-MB a lot because as a general rule I’m not a fan of overall compression. The Pro-MB allows me to control the dynamics on specific frequency bands, and means I often don’t have to use a classic compressor at all. I’ve created a default preset for the Pro-MB that I like to use as a starting point, in oversampling mode, and I then tweak it from there. I’ll also use the iZotope Exciter at times, to inject upper and lower mids. But in general I prefer to use saturation and limiting instead of compression, particularly because many of the tracks that I’m given to mix these days use samples that have already been compressed and completely levelled, so on these you’d only use compression as an effect or to add character, and I prefer saturation for that. Vocals that I get from other producers also tend to have been compressed by them, so I don’t necessarily have to do it again.”