Even if your sampler has 128 MB of RAM or more, sometimes you can't or don't want to use stereo samples. Stereo samples use at least two voices of polyphony per note-more if you're using multiple crossfaded layers. A stereo instrument with three velocity-crossfaded layers consumes 6 voices per note; a five-note chord eats 30 voices. If the instrument has a long release time or you repeatedly use the sustain pedal, you can easily have 90 or more notes going at one time. Even if your sampler has 128 voices of polyphony, you're left with considerably less room for other instruments, especially if they, too, are in stereo.
In addition, stereo samples use twice as much sample RAM as mono samples of the same instrument. This isn't a big problem in the studio because you load an instrument into RAM only when you're ready to lay down its track. But in a live performance, you may need to have hundreds of different instruments ready to go on demand.
What are your options if you like a dramatic stereo effect but can't afford the resources that you need to use stereo samples for every track? And what if your samples are available only in mono, like those in ROM-based synths and samplers?
EXPANDING YOUR WORLDFortunately, there are several ways to create a satisfactory artificial stereo image from a mono sample. But will it really be satisfactory? Yes, sometimes. Your ears and brain can be very picky about the precise nature of a stereo image. For instance, a listener is more likely to detect an artificial stereo image of a single instrument. But when it comes to simulating a pleasant stereo ambience, you can often just "throw things together," and the brain will quite happily interpret the results as a nice, wide stereo field.
Each method of generating a stereo sample from a mono source has good and bad points. I'll discuss some of the more common methods, then share a very successful technique that I developed for some of my CD-ROM projects.
Although this article focuses primarily on techniques to be used with samplers and ROM sample-playback synthesizers, many of the tips are equally useful for any type of monaural audio source. For example, you can try some of the tricks on a monaural live recording or mono audio track from a video.
BACKWARD COMPATIBILITYLet's consider some possible consequences of using artificial stereo images. Our pseudostereo sounds will be wonderful when played on stereo systems, but what if, after all that work, they are played back in (shudder) mono? They still need to sound good, of course, and achieving that is not always simple.
Why would they be played in mono? AM radio stations and many house P.A. systems in clubs and arenas are mono. In addition, certain effects processors-both external units and onboard effects in synthesizers and samplers-sum the left and right inputs to mono before they apply a stereo effect. The odds are good that your beautifully crafted stereo mix (or stereo sample) will get played in mono somewhere along the line. So when you're creating a stereo image, it pays to consider mono compatibility.
This is easier said than done. Unwanted sonic artifacts often appear when an artificial stereo image is collapsed down to a monaural channel. One common artifact is comb filtering, which gives the sound an unnatural, tubelike character or creates annoying peaks or notches in specific frequency bands. Another possible problem is the loss of clarity, or even a complete cancellation of the signal. So when you create your pseudostereo image, you must be very sly.
THE USUAL SUSPECTSHere's a quick roundup of seven common image-manipulation techniques for making a stereo sample from a mono source. I'll note the primary benefits and drawbacks of each method.
Panning frequency bands. This is one of the more delicate (that is, less intrusive) ways to expand an image. Take a mono instrument sample and separate it into four or more frequency bands. Pan each band, alternating between the left and right sides. Try all panning values for each frequency band, and experiment with the center frequencies and bandwidths if those parameters are available. Make sure you occasionally pan your mixer to mono to hear what the left and right channels will sound like when blended.
This effect works best if you can use four or more bands. (With only three bands, the highs and lows would be panned left, leaving only the mids to go to the right.) If you do this properly, the stereo illusion's quality will be determined mostly by the bandpass filter's phase accuracy and control over its bandwidth.
Delaying or detuning one side. Two of the most common methods of creating an artificial stereo image on synthesizers and samplers are delaying a copy of a sample relative to the original version and detuning a copy relative to the original. You can also combine the two techniques.
There are several ways to use delay. One method is to assign separate but identical instrument layers to the left and right channels. Delay playback of one side via the sampler's Start Delay parameter, then pan the channels to opposite sides.
Another method is to split the mono signal into two identical signals, delay one signal relative to the other, and pan the dry (unprocessed) and wet (delayed) signals to opposite sides. One common way to do this is to send the mono signal through a digital delay line (DDL) set to 100 percent wet (processed). If the DDL has separate dry and wet outputs, you can split the mono signal that way. Otherwise, you'll have to split the signal before the delay by, for instance, sending the same mono sample to the left and right synth outputs.
You can also use a sample-editing program. Transfer the sample to your computer and open the mono original on the left track of a stereo file. Copy the sample to the right track. Trim or insert silence at the beginning of the copy, phase-shifting it relative to the original (see Fig. 1). Save the file for future use, and fly the new stereo sound back into your sampler.
All of these approaches accomplish the same thing: they delay playback of a copy of the sample. You just have to decide whether you prefer to sacrifice additional polyphony to the second layer in your sampler, use a signal processor (the delay), or muck about with editing on a computer.
Although popular, the delay method is not without flaws. If you're using a start delay (or delay time) between 2 and 20 milliseconds, you could end up with comb-filtering artifacts when the signals are combined to mono. Delay times longer than 24 milliseconds tend to result in fewer comb-filtering artifacts, but the signal will increasingly sound like a rapid slapback echo as you lengthen the delay time.
In addition to adjusting the time difference between the left and right sides, you can slightly detune one side relative to the other. Usually you do this in the Pitch or Oscillator page of your synth or sampler. You can subtly detune by raising or lowering the pitch of one side by one or two cents. If you want a more noticeable offset, raise the pitch on one side and decrease the pitch on the other side by the same amount. This will help keep the overall pitch centered instead of sharp or flat.
For an even more animated quality, assign an LFO to the pitch of one or both sides. Again, by making the pitch respond positively to the LFO on one side and negatively on the other, you'll keep the average pitch centered.
Phase inversion. Of all the techniques used to create a stereo image, this produces the most unnatural artifacts. As with the "delay one side" trick, you can do phase inversion with two instrument layers or with a digital delay that allows you to invert the phase of the processed signal while keeping the delay time at 0. Once again, you'll pan the dry and wet signals to opposite sides. Two considerations here are whether you can afford to use twice as much polyphony and whether you have a processing block that can invert the signal phase.
The end result is a dramatic, hyperwide stereo image that, depending on your point of view, is either radically unnerving or sonically irritating. When you listen through headphones or sit right in the sweet spot, you'll feel as if the insides of your ears are being pulled away from your head. And you can forget about summing to mono. If the two sides are truly 180 degrees out of phase, the mono version will be silent because the two sides will cancel each other out.
Different sample-start points. If you have a long, ambient, textural sample, or a sample of a complex analog synth oscillator, try this trick. Create a copy of your ambient sample (such as the sound of rain, a bubbling stream, or a vinyl record spinning around) and make a new start point one-third to one-half of the way between the original start point and the loop point. Now, when you press a key, one side of the stereo signal will start at the beginning of the ambience and the other will start somewhere in the middle. Experiment with panning the two sides, and you'll come up with a lively and highly dimensional sound that remains interesting even after repeated loops.
By using this technique on a sample of a thick analog oscillator, you can create a huge sound with subtle and intriguing stereo motion. Try it on a sample of white noise-you'll perceive a very complex stereo image that doesn't seem to repeat.
If your synth or sampler can choose sample-start points at random, you can get even more interesting results. Again, experiment with wide and narrow panning. (If the panning is too wide, the sound will seem unnatural.) When you find the right spot, you'll hear elements that seem to exist at points all over the stereo field.
Different sample-loop points. This is similar to using different sample-start points, but the stereo perspective changes while the samples are looping. Simply make a copy of the original sample and find somewhat shorter loop points, so that when the shorter loop is hard-panned against the original sample, you have a complex stereo interaction that plays for a long time before the cycle repeats. The use of different start and loop points usually translates well even if the two sides are panned all the way back to mono.
Different start and loop points are effective on a sample of white noise because, by definition, the spectral content of white noise is random. The fact that a 2-channel sample of white noise repeats itself can be disguised by the interaction between the left and right signals. You can then filter this stereo version to create the ambient sound of thunder, wind, waves, rain, and other noise-based sound effects.
Different instrument samples. This technique is most successful with ensemble samples. Let's say you have two different sax-ensemble synth instruments. By using two layers and carefully experimenting with each layer's panning, volume, and tuning, you can achieve a believable and lush stereo ensemble that is far more pleasing than either ensemble alone.
The key is to keep trying different values for each of the parameters until you find the sweet spot. The values may be different for each keymapped sample range. Use your ears to create a uniform ambience all the way up and down the keyboard. A stereo image generated in this way can almost always be collapsed to mono with minimal artifacts.
Short reverbs and room effects. You can create a full-sounding stereo image by using either a "short" room setting on a room simulator or just the early reflections parameter on a digital reverb. But you're likely to hear artifacts if the signal is combined back to mono.
MANUAL SHIFTINGManual timbre-shifting is not a stereo-imaging trick. It is a method of altering the playback of a monaural, multisampled instrument in such a way that the edited instrument exhibits timbral characteristics that are shifted in both pitch and time; the end result is not an artificial stereo instrument but rather an enhanced monaural instrument. The process involves remapping the multisamples up or down to the next keymap range and then modifying their playback rate so that the samples play the correct pitch. Some samplers and sample-playback synths have a dedicated parameter for doing this, whereas others accomplish it through several related parameters. Unfortunately, some synths won't let you pull off this trick at all.
Here's how it works. Let's say you have a piano multisample with a different sample for every whole step. You might have a sample at C4 and another at D4, E4, F#4, G#4, A#4, C5, and so on, up and down the keyboard.
Manually timbre-shift this instrument by transposing it up a whole step and then lowering the pitch by 200 cents (or a whole step) in your synth's Pitch page (see Fig. 2). Now, pressing the C4 key actually plays the D4 piano sample, but because you decreased the pitch by 200 cents, you'll still hear the correct C note. However, you have only changed the lowered sample's frequency, not its formant or time base, so it will sound slower, deeper, and heavier than normal. Conversely, if you transposed the piano down a whole step and then raised the pitch by 200 cents, the result would be a piano with a slightly faster, brighter, and thinner sound.
To create an artificial stereo ambience with this timbre-shifting technique, simply pan an unaltered version (that is, playing C4 triggers a real, unshifted C4 sample) to one side and a timbre-shifted version to the other side. By experimenting with upward and downward timbre-shifting, the number of half steps that the instrument is shifted, and the depth of panning, you can get a wide variety of stereo images that also contain a much fuller sound.
Timbre-shifting does have drawbacks, though. For one thing, you need to tune each sample's pitch very carefully or the two sides will chorus. Furthermore, the instrument's pitch must be stable over time, because differences between the left and right layers will be very apparent. Finally, many instruments do not have the same number of notes in every key range. For example, an instrument might be sampled every six notes in the bottom range, every two notes in the midrange, and every octave in the extreme high range. Thus, timbre-shifted notes may act differently in the various ranges, leading to an uneven stereo image across the keyboard.
HANDCRAFTED BEAUTYHandcrafted stereo timbre-shifting is a method that I developed over the past five years while programming roughly 24 sample CD-ROMs. I put the technique to full use on a guitar CD-ROM that includes 115 different acoustic and electric guitars and basses. All of the guitar samples that were originally recorded in mono also feature stereo versions that you can sum to mono or pan to any desired position in between without generating artifacts.
The process is based on the timbre-shifting method described earlier, but instead of simply shifting a mono sample, you create two identical layers (as with some of the other techniques I've discussed) and shift the timbre for one or both channels. After that, you can individually fine-tune the samples in every key range, or even every key. The resulting artificial stereo field is pleasant and consistent over the whole keyboard. The technique works equally well on ROM- and RAM-based sampled instruments, but for the best results, your instrument multisample should have more than two samples per octave so that the samples won't get shifted too far from their normal pitch range.
SHIFTY STRINGSLet's start with a string ensemble that has been sampled three times per octave, with the sample roots at every C, E, and G# throughout the instruments' range. For now, we'll assume that the root keys are at the bottom of each key range, so every C sample is pitch-shifted to cover the range from C to D#, every E sample covers E to G, and every G# sample handles G# to B.
Make a copy of the keymap; name one copy String Ens L and the other String Ens R. String Ens L will be in our program's first layer, panned hard left. String Ens R will be in the second layer, panned hard right.
Next, edit String Ens L, starting with the first sample key range. Let's say that the first sample is StrEns C2 and has a key range of C2 to D#2. Instead of using the C2 sample, dial in the next sample above it, StrEns E2. (Remember, the sample roots are at C, E, and G#.)
Now the keys between C2 and D#2 trigger the sample rooted at E2, so if you play any key from C2 to D#2, you will find that the edited String Ens L keymap is playing E2, F2, F#2, or G2 (see Fig. 3). Use the coarse-tuning adjustment to bring the pitch back down a major third so that String Ens L's pitch is the same as the pitch of the unedited String Ens R sample. Make the stereo signal beat free with the String Ens L keymap's fine-tuning parameter.
When the layers' tuning matches as closely as possible, listen for left/right volume imbalances. On the left keymap, use the Volume Adjust parameter to create a perfectly centered image.
Go through the String Ens L keymap and change each sample to the one that's directly above it. Then use the coarse-tuning, fine-tuning, and volume controls to create a centered image and tuning for each sample.
When you're done, you'll have a lush and spatially animated stereo string ensemble. Because the left and right samples for each key range are different, you can pan each layer anywhere from hard left or right all the way back to center without creating annoying artifacts. And the pair of keymaps will sound good through digital effects that sum their inputs to mono.
THE FLIP SIDEOur string-ensemble example paired a normal keymap with a keymap of higher samples that were shifted down. This results in a full, slow, thick sound. You'll get the exact opposite results if you use a normal keymap in the left side and modify the right side. This time, start at the highest key range and dial in the sample below it.
For example, the string ensemble was sampled every fourth semitone (or three times per octave). If the top key range has a C7, replace it with the G#6 sample (the next sample down); where the G#6 sample was, put an E6; and so on down the keymap. Again, tune and adjust volume for a nicely centered image.
FURTHER EXPLORATIONSYour new stereo instrument has a bright, snappy crispness. But don't stop there. Try extreme timbre-shifting in opposite directions for each side and listen to the results. If you don't care about polyphony, add another nonshifted layer and pan it center. Then pan the two timbre-shifted layers hard left and right. You'll get fatness and clarity at the same time!
I hope this exploration of creating artificial stereo images gives you some ideas of your own. The handcrafted timbre-shifting method takes a bit of time and focus, but the results can be pure sonic satisfaction.
Daniel Fisher is the director of soundware engineering at Sweetwater Sound. He has designed factory and third-party soundware for Alesis, E-mu, Korg, Kurzweil, TC Electronic, and Yamaha. But he really wants to be a guitar player.