Hannes Keseberg: our engineer helps an artist fake a big-ensemble sound by layering recordings of a string quintet.
The quality of virtual instruments these days makes it pretty easy to get a satisfying string-ensemble sound most of the time. Even the best libraries, however, still struggle with more expressive and melodic parts on those occasions when strings are heard in isolation — not just because of technical limitations of the software, but also because creating convincing parts requires both serious arrangement experience and fastidious MIDI programming. In short, there are times when only real instrumentalists will cut the mustard.
One artist who recently ran up against this was Hannes Keseberg (https://en.hanneskeseberg.com), who asked me for advice on some of his work-in-progress productions, one of which had a 45-second introduction comprising just string orchestra and lead voice. The intention had clearly been to achieve a classic, lush string sound to match the crooning vocal style, but neither Hannes nor his musical collaborator Matthias were satisfied with the results of their own attempts to simulate this kind of large-scale ensemble sound artificially. Not that their approach itself was ill-advised. I’d likely have followed a similar route myself under the circumstances, creating a synthetic large-ensemble sound with a virtual instrument and then recording individual instrument overdubs over the top to improve realism. The devil’s in the detail, though, and here a combination of slightly awkward programming and thin-sounding close-miked overdubs had fallen short of generating a convincing live-ensemble illusion. So I recommended they take a punt on live strings instead.
Mind you, hiring and recording a full string orchestra commercially was out of the question on financial grounds, so I suggested that he get a smaller group into a nice-sounding room, and then layer up several takes of them to fake a larger ensemble. Based on the outcome of their previous string recordings, Hannes wasn’t convinced that they’d be able to get a sufficiently usable outcome, so I also agreed to lend them a hand with the tracking session itself to help them get a more solid sound.
Fortunately, Hannes isn’t short of contacts in the local music scene, so he was able to get us into a local college recital room with the Besamim Quartet (www.besamim.de), and an additional double-bass player to help fill out the low end of the desired orchestral texture. Although the target sound was going to be fairly traditional, I did recommend a slightly less orthodox ensemble layout for recording purposes, moving the double-bass and cello to the middle of the group to keep the low frequencies centred, and then placing the second violin over on the right-hand side to balance the first violin in the stereo picture. This was because I knew the string introduction preceded an otherwise fairly modern-sounding pop arrangement, in which context central bass and a fairly balanced stereo picture tend to be desirable.
Beyond that concession to modernity, though, I deliberately used a miking setup similar to that of an orchestral scoring session: a principal stereo rig above the centre of the ensemble supplemented by closer, per-instrument ‘spot’ mics and a pair of more distant room mics. In this kind of configuration, the idea is for the main stereo pair to be far enough away from the instruments to pick up a fairly natural-sounding representation of them, and in an ideal world those two mics might even give you everything you need. What all the ancillary mics do is provide something of an insurance policy in case you need to change the wet/dry ratio or internal ensemble balance at mixdown. Things sounding a bit too dry? Fade in some of the room mics. Viola too quiet? (Is that possible?) Fade up that player’s close mic. This fall-back was particularly important, given that we had to set up our recording equipment in the recital room itself, which meant monitoring on headphones — the stereo-stretched ‘in your head’ listening experience of cans tends to make reverb-level judgements in particular rather unreliable.
When working on a budget, I find that using omnis for my main stereo pickup usually gives the best results on full-spectrum acoustic ensembles like this, especially at the low end, so I decided to use the best omnis I had available (Shure KSM141s) as a spaced pair above and in front of the ensemble. Immediately upon entering the recital room, however, I began to have some reservations about my plan. While the room had clearly benefited from some acoustic design, its sheer capacity (an eight-metre ceiling height and space for a couple of hundred seats) gave substantially more reverberation than I’d hoped for. Otherwise it sounded great, though, so I decided just to bring my mics in a bit closer and see how we got on.
Despite factoring this in, my first miking-position guess (just beyond the front edge of the stage where the string players had set up) was still too roomy. I’d taken the precaution of putting my main pair onto a big cathedral stand, so I was able to extend its boom arm to bring the rig further over the centre of the group without difficulty. This gave an immediate improvement in definition, but seemed to leave the stereo image feeling overstretched. Stereo width is something that’s also extremely difficult to judge over headphones, so I first corroborated my impressions using Melda’s MStereoScope vectorscope plug-in (in conjunction with a commercial reference recording I’d brought with me), before deciding to narrow my initial 60cm inter-mic spacing to 45cm — widening the rig’s acceptance angle and thereby counteracting the worst of the image stretch.
By this point the main pair was sounding pretty respectable to me, so knowing that I didn’t have tremendous amounts of setup time on my hands (when do I ever?) I began cracking on with the ‘insurance’ mics. Given that the overall stereo pickup so far was still a touch on the roomy side, I chose to get the close mics going first. My first was a single Brauner Valvet large-diaphragm valve mic we’d been able to borrow for the session, and which I’d decided to use for the bass. A lot of small-studio engineers instinctively use spot mics with directional pickup characteristics, but I left this mic in omni because of the truer and more extended low-frequency response typical of this polar pattern. “But what about spill?”, I hear you cry! Well, yes you do get a bit more spill with an omni, but even at the distance I had it (about two feet from the instrument), the bass’s acoustic power relative to the higher strings allowed it to dominate enough in the spot-mic’s balance that the resulting signal remained useful for balancing purposes — which is the main point of a spot mic. Furthermore, indirect spill on an omni mic, while generally dulled by off-axis high-frequency insensitivity, nevertheless tends to stay fairly natural-sounding, so it tends to be less timbrally damaging, in my view, than the lesser level of more unnaturally coloured off-axis spill you’d typically get from a cardioid. Finding a suitable miking position was then primarily a question of listening around the instrument for an appealing tone, and putting the mic there! That said, the position we ended up with — around the level of the bridge and slightly off to one side — does quite frequently seem to deliver the goods for me in practice, so it’s often a good starting point if session time is of the essence.
One of the most frequent problems people encounter recording strings on a budget is a thin and scrappy timbre. While weak players and mediocre instruments are often partly to blame (although clearly not in this specific case!), using cheap condensers for spot-miking is also a critical factor, I think. I’ve tried a lot of different affordable condenser mics (both large- and small-diaphragm models) on strings, and frankly many of them sound pretty hideous at typical close-miking distances, exposing high-frequency harshness and mechanical noises in the most unflattering manner. As a result, I now prefer to use ribbon mics instead if I can, and on this session was able to cover all the remaining instruments in this way: Matthias’s Beyerdynamic M130 for the first violin and my own three trusty bargain-basement Superlux R102s for the second violin, viola, and cello.
Another common failing of project-studio string recordings is spot mics being placed too close, in an effort to minimise spill at all cost. It’s crucial to realise that it takes a bit of distance for the different frequencies generated by string instruments to blend sensibly, and miking too close also gives unattactive bow scrapes insufficient distance to dissipate, so I’d almost always prefer to sacrifice a bit of separation in pursuit of a nicer basic timbre. The figure-of-eight ribbon mics played into my hands a bit here, not only because their inherently warm sound avoided abrasive high-frequencies, but also because they inherently have a pretty tidy off-axis response, so I was fairly confident any spill they did pick up wouldn’t sound too unpleasant.
The ribbon mic placements did take a fair bit of refinement, though, and not just in search of a useful general tone. Further tweaks were required to achieve a useful degree of separation on each mic — the signal won’t be much use to you at mixdown if the sound of the instrument the mic’s pointing at is quieter than the spill! Again, the figure-of-eight polar pattern was a big help here in rejecting spill from neighbouring instruments without having to move any of the mics in closer than about two feet. As with any multi-mic setup, though, I was careful to experiment with the polarity relationships between the mics to avoid any catastrophic phase cancellations in the monitoring signal.
A final step was to put up some room mics: a pair of Neumann KM184 small-diaphragm cardioids from Matthias, placed eight feet apart in the second row of audience seats and pointing away from the stage. Since I was pretty sure the main pair were already roomy enough, I deliberately chose the wide room-mic spacing to produce a super-wide stereo impression, giving me the option to use a little of the room signal as a possible width-enhancer later on if I wanted. Some engineers like to spread the ensemble itself across the whole stereo width, whereas I prefer to leave a little space at each stereo extreme for the room reverb to paint out to the edge. Within this context, a little extra width to the room reverb can be a nice feature.
Although I managed most of the setup while the group were rehearsing, I did still ask them to play the piece a couple of extra times just for me, so I could confirm that the mics were indeed delivering what I was looking for. I normally do my best to impact on the performers as little as possible while recording (beyond clambering around them with mics, obviously!), which means I always endeavour to hit record when the players are ready, whether I’m fully ready or not. In this case, though, I knew that the performers were all highly trained classical musicians, and that the sonics were as important as the performance here, so I figured it would be a false economy to start doing takes while any question marks still hung over the miking setup. Fortunately, the final tweaks didn’t take more than about 10 minutes, so we were soon laying down tracks.
If you’re going to layer up multiple passes of any instrument, you need some means of synchronising later performances to previous takes. Where you’re overdubbing over an existing arrangement this is rarely tricky, because players can navigate by rhythmic events in other musical parts — the drums, for instance. Where such rhythmic events don’t exist in the arrangement, a click track can bail you out as well, as long as you make sure that its sound in the overdubbing performers’ headphones doesn’t audibly bleed into the microphones. In this instance, neither option was much use to us, because the strings in this introduction were both unaccompanied and rubato. Furthermore, Matthias felt that the flow of the performance should be led by the string players, and didn’t want to cramp their style by laying down a click track in advance of the session and forcing them to use headphones. So we decided to try recording one take, and then synchronising all subsequent performances to that.
What complicated matters was that there wasn’t much in the way of rhythmic information in the string arrangement, so it was clear to me that we’d have to generate some kind of rhythmic guide track on the session if we were to avoid anyone dying of tedium at the editing stage. Every DAW now has facilities for stretching its internal tempo map around a free-running audio recording, and this allows you to generate an appropriate click track for most eventualities fairly swiftly, but I deliberately steered clear of this because Matthias was recording in Pro Tools, and I wasn’t convinced that either of us were familiar enough with its tempo-mapping facilities to avoid the horror of brow-furrowed manual-surfing under the scrutiny of five impatient musicians. (Believe me — if you’ve ever been there before, you never want to go there again!)
So I suggested we do things a bit more old-school, asking Matthias to grab an SM58 from his kit bag and lay down his own guide track (basically him rapping numbers!) alongside the live strings take we already had. Although this is inevitably a bit of a trial-and-error process, it’s actually surprisingly quick and intuitive to do, so the musicians had scarcely had a chance to recover from my collection of elephant jokes before we were ready to get going again — this time with Matthias conducting them while listening to his own recorded directions over headphones.
Layering up the rest of the takes was straightforward after that — classical musicians are nothing if not consistent! To help the final orchestral illusion, I also suggested we do some additional passes where all the musicians moved their chairs a metre or so further from the mics. Although this wider ensemble spacing would have caused problems with the musicians’ natural communication if they’d been playing on their own, the presence of Matthias directing helped keep everyone well enough together for our purposes.
Returning to my mix room with the files, I was gratified to find that the raw recordings still held up pretty well on my monitor speakers. As I’d suspected, the main mics were a little too roomy, so I ignored the room-mic channels to start with and judiciously faded in the spot-mic signals to sharpen up the focus to taste. There’s a natural tendency for low-end spill to build up when using spot mics, and our main mics already had plenty of low-end warmth to them, so the ribbon mics all benefitted from high-pass filtering, but otherwise there was precious little I felt I needed to do get a good quintet mix sound — just a few decibels of high-frequency lift to brighten the restrained high-end of the ribbon mics a fraction. It’s worth stressing that I didn’t feel the need to cut any harshness at all at the high end, which is something I’ve never been able to manage working exclusively with condensers. Further experimentation with the polarity relationships of the mics seemed to improve the sound too, now that I had the luxury of time and a more reliable monitoring environment in which to engage with the smaller niceties, and I even decided to sneak in a little of the room mics in the end, for their extra stereo width, albeit high-pass filtered to avoid blurring the low end.
For the large-ensemble illusion, I ended up layering three regular passes with two of the more distant passes, the latter with their bass close-mic tracks muted — most string orchestras have fewer basses than violins, after all, and this seemed to make the final effect more convincing. Speaking of which, how believable was the end product? The simplest answer to that question is: judge for yourself! There’s a full selection of audio demonstration files to accompany this month’s article at https://sosm.ag/apr16media, including raw files from all the mics for one of the close passes, as well as a full ‘fake orchestra’ mixdown.
The challenge of building a big ensemble string sound on a budget is something many project-studio engineers come up against, so it’s no surprise that it’s appeared in our Session Notes column before — back in SOS April 2014, to be specific. If you missed that article, do track it down in the SOS archives, because author Neil Rogers clearly explains why he took a different approach from mine, spacing players further apart and favouring the close mics more to allow greater post-processing flexibility. There are many ways to skin this particular cat, so it pays to evaluate tactics from a range of different engineers if you’re still deciding on your own preferred strategy.
It’s always the same — you get back from a session, and end up kicking yourself that you missed some trick or other that could have helped. In this case, it occurred to me afterwards that we could probably have improved our ensemble illusion if we had persuaded the first and second violinists to swap instruments (or else their seating positions and parts) to reduce the sonic similarity between the different passes. As it stands, my biggest concern with final sound is a slight whiff of chorusing, even though there’s no chorus effect in the mix. It’s pretty subtle, admittedly, but enough to niggle me into remembering that dodge next time...
The audio files available at https://sosm.ag/apr16media accompany my Session Notes article in SOS April 2016 about simulating a big-ensemble string sound by layering recorded takes of a string quintet. The filenames are fairly self-explanatory, but here are some additional notes to describe exactly what you’re hearing in each case. (Incidentally, if you’d like to try mixing the basic string-quintet sound for yourself, just import all the files with the ‘_Raw’ filename suffix into your DAW software at the same starting position and you should be good to go!)
Here’s how the main stereo pair of Shure KSM141 small-diaphragm omni microphones sounded by the time I’d finished shifting it around to my satisfaction. With hindsight, it’s still a touch on the roomy side, but this aspect of the capture wasn’t easy to judge on the session, given that we had to monitor on headphones in the recording room. Fortunately, we’d hedged our bets in this respect by recording a full complement of close microphones, allowing us to introduce more direct sound as required for each individual instrument.
The double-bass spot microphone was a Brauner Valvet large-diaphragm valve microphone operating in its omni mode. I chose this polar pattern primarily for its excellent low-frequency extension, and despite its notionally lack of directionality, it actually picked up a respectable direct/spill balance even at a miking distance of around 60cm from the instrument. Furthermore, the spill it does pick up retains a fairly natural timbre (albeit somewhat dulled by the mic’s off-axis high-frequency insensitivity), which helped retain a representative tone and balance for the ensemble as a whole.
I used ribbon mics for all the other string parts, and selected the highest-quality of these (a Beyerdynamic M130) for the first violin. The mic was about 75cm above and in front of the instrument, pointing downwards towards the soundholes. Once a promising timbre had been found, the figure-eight polar pattern’s null plane was angled towards the neighbouring viola player to improve separation. In this audio example you can hear the raw output from this microphone. Notice how the ribbon mic’s naturally rounded tone prevents the violin from becoming excessively biting at this comparatively close miking location, even when the player moves onto the ‘E’ string at 0:13, and also how spill from the rest of the ensemble is again presented very naturally on account of the well-behaved off-axis response of the figure-eight polar pattern.
All the spot-mic signals other than the double-bass were high-pass filtered at the mix to reduce low-frequency spill and focus their timbre a little towards the mid-frequency definition that was lacking in the main stereo pair’s sound. In the specific case of the first violin, this filter was from FabFilter’s excellent Pro-Q parametric equaliser plug-in, turning over at 150Hz with a slope of 12dB/octave and a moderate Q value of 0.7. This audio demonstration shows what that frequency adjustment sounded like. It’s not an enormous difference, so you might want to compare it with the 03_ViolinSpot_Raw file directly in your DAW to hear it more clearly.
The second violin was miked with a Superlux R102 active ribbon microphone placed about 60cm above and in front of the instrument. The microphone’s rejection null was then directed towards the cellist sitting just past the player’s right elbow. As with the more expensive Beyerdynamic ribbon mic on the first violin, the off-axis tone of the Superlux’s figure-eight polar pattern again renders spill from the rest of the ensemble very naturally.
At the mix, the second-violin spot mic was processed the same way the first-violin’s was: with a 12dB/octave high-pass filter at a Q value of 0.71. Again, this impacted very little on the violin’s own timbre, as you can hear if you compare this example directly against the preceding 05_Violin2Spot_Raw file.
The viola was captured via another Superlux R102 active ribbon microphone, again placed above and in front of the instrument, although in this instance a little closer (more like 40cm away) in order to achieve sufficient separation for the viola’s direct sound — the violin and bass were otherwise overpowering the viola in the close-mic balance, even after careful orientation of the ribbon mic’s rejection plane.
As with the other ribbon spot mics, high-pass filtering was used to avoid a build-up of low-frequency spill. For this track the cutoff frequency was set slightly lower than for the violins, at 120Hz, but with the same 12dB/octave slope and Q value of 0.7. Given that the lowest fundamental frequency a viola puts out is around 131 Hz, it’s hardly surprising that the processing doesn’t dramatically affect the instrument’s sound character.
Here’s the raw cello recording, captured by my third Superlux R102 active ribbon microphone. The mic was about 60cm from the instrument roughly at the player’s head height and pointing downwards towards the middle of the fingerboard. The microphone’s null plane did a very good job in this instance of rejecting the other members of the ensemble, although some significant bass spill is still apparent.
I set the cutoff point for the cello’s high-pass filter to 100Hz, which not only removed a good deal of bass spill from the signal, but also significantly recessed the cello’s lower fundamentals, simply because there already felt to me to be plenty of low-end weight coming from the main stereo pair (as you heard in the previous 01_MainPair_Raw audio file). The filter’s slope was again set to 12dB/octave with a Q value of 0.7.
This audio example demonstrates the combined sound of all the spot mics, each panned to roughly to match its instrument’s position onstage. While experimenting with polarity settings for each track, I preferred the overall sound with the second violin and cello mics polarity-inverted. Beyond this, however, no further processing has been applied beyond the individual high-pass filters on each of the four ribbon mics.
The only further EQ processing I applied to the spot microphones for mixdown purposes was a global high-frequency boost. Ribbon mics are by nature rather restrained at the high end, and I felt that they needed to be a little more forward-sounding on the whole within a modern production context. It wasn’t a big deal though — just a simple 4dB boost using a gentle Q=0.15 high shelf set at 9kHz. Directly compare this demonstration with the 11_Quintet_AllSpots file in your DAW to hear the difference most clearly.
Combining the processed spot mics with the main stereo pair of omni microphones now gives us the sound you can hear in this audio example. Comparing this mix with the 01_MainPair_Raw file, you can hear how much the spot mics help focus the individual instrument sounds. Comparing against the 12_Quintet_AllSpotsHFLift file, on the other hand, demonstrates how much life and space is provided by the main stereo pair’s more holistic pickup. Although I was pretty happy with this sound already, I thought I’d experiment to see whether the room mics might add anything further of merit.
Here is the raw room mics recording. As you’d expect of a spaced-pair stereo rig with an eight-foot spacing, the sound is super wide. I quite like recording room mics this way, because it allows me to use them to create width without having to stretch the ensemble itself too much across the stereo image. However, if I’d tried to add this signal to the mix you hear in the 13_QuintetMainsPlusSpots audio file, I’d have quickly started blurring the low end.
Tightening up the low end of the raw room mics signal you heard in the 14_RoomMics_Raw audio file was a simple matter of gently rolling off the low-frequencies from about 200Hz with a gentle (Q=0.25) 12dB/octave high-pass filter — this time using the ReaEQ parametric equaliser plug-in bundled with Cockos’s Reaper DAW software. To hear how this signal functions in context, check out the following 16_Quintet_FullMix audio file.
Here’s my full mix of the string quintet, comprising signals from the main stereo pair, the five spot mics, and the pair of room mics. I’ve also put a little level automation onto the spot-mic signals to clarify the instrument balance at a few points.
Combining three takes of the music heard in the 16_Quintet_FullMix audio file results in the following ‘fake’ ensemble sound.
In addition to the normal takes, we also recorded some passes with the players sitting further away from our mic rig. Here’s a mix of a couple of those takes, so you can hear how the position change affected the sound. I’ve also muted the bass spot mic for these takes, the idea being that when they’re later added to the close-position takes heard in the 17_Orchestra_FullMixCloseLayers audio file, you’ll get more of the high-strings bias you get in a real-world string orchestra, where the violins, violas, and cellos typically outnumber the double-basses.
And finally, here’s the complete simulated large-ensemble string sound, combining three close-position takes and two distant-position takes. In total, this mix uses very minimal processing — a total of five high-pass filters, one high shelving EQ boost, and two polarity-inversion switches — so you’re mostly hearing just what came out of the mics themselves.