(This column originally appeared in the March 1978 issue of Contemporary Keyboard magazine.)
Last month's column showed the frequency and phase response graphs of a multiple resonance filter array. Shortly after I wrote that column, an oscilloscope camera found its way into our lab. I took some pictures of the waveform produced by our experimental multiple resonance filter array while it was being fed with a sawtooth wave. I think these photos shed some light on why violin tones sound the way they do and why, with acoustic timbres, there is really no such thing as "the waveform of a tone color."
Our experimental multiple resonance filter array has 24 resonant sections. The resonance frequencies start around 100Hz (for good 'cello tones), and go up to about 8kHz. To display the waveforms, we fed a 250Hz sawtooth wave into the filter array's input. One cycle of the input waveform is shown in Fig. 1. The filter array's output is shown in Fig. 2. The output waveform is a complex, irregular succession of ups and downs, quite far removed from the original sawtooth. Fig. 3 is the same as Fig. 2, except that the frequency of the sawtooth input was lowered by 1% (2.5Hz). That's the amount of frequency change that might occur during a shallow vibrato. You can see that the waveform has many subtle changes, but the overall contour is only slightly modified. Now look at Fig. 4. The frequency of the sawtooth input has now been lowered by a semitone (6%, or 15Hz), and the waveform's contour has undergone major changes. Only the general character of the shaping remains.
To the listener, the waveforms shown in Figs. 2, 3, and 4 all have the same timbre. Furthermore, the timbre seems most violin-like when the pitch glides from one note to another, or when it is animated as in vibrato. This means that not only is there no such thing as a unique "violin waveform," but that a waveform that changes dramatically with frequency is essential for convincing string tone simulation. This is true not only of string timbres, but of brass and woodwind timbres as well. The most straightforward way of achieving this sort of waveform-versus-frequency dependence in electronic instruments is to employ complex filters that are modeled after the resonant properties of acoustic instruments.
Faking String Instrument Frequency Response
Unless you're unusually well equipped (hardware-wise, that is), your electronic keyboards won't have complex filters of the type needed to simulate really good string timbres. Two ways of faking it are to build up a complex frequency response with several simple filters or to modify a spring reverb unit. Let's take them one at a time.
To build up a complex frequency response, you should have a simple mixer and a variety of linear (non-distorting) sound modifiers. Graphic and parametric equalizers are good, and wah-wah pedals can be used, as can phasers that allow you to shut off the modulation. Feed your keyboard signal to all the modifiers simultaneously, using jack multiples or Y-cords. Feed the outputs of the sound modifiers to the mixer input. Play with the controls, including the equalizer controls on the mixer. I won't guarantee that you'll wind up with a fat string sound, but I have heard this approach work more than once.
If you have access to a spring reverb, try experimenting with stuffing the springs themselves with a very small amount of fiberglass or foam rubber. This absorbs some of the vibration in the springs, and therefore shortens the reverb time. If you get the right amount of sound-deadening material in the springs, the reverb unit will lose all of its metallic twang and begin to sound a bit woody. This is because the frequency response of a spring reverb consists of a series of very sharp, closely spaced peaks. Damping the vibrations in the spring spreads the peaks out and reduces their heights.
One approach to string timbre fakery that probably won't work is the use of a flanger. It is true that a flanger's frequency response consists of a series of peaks and dips. However, these peaks and dips are harmonically related, which means that a note's harmonics will either miss all the peaks, or they will hit a whole series of peaks. This produces an unviolin-like response in which certain notes will tend to jump out. A well-designed violin body or multiple resonance filter array has peaks that are not harmonically related, thus emphasizing some harmonics and attenuating others of virtually any pitch that you care to feed into it.
To simulate string tones, you should start off with a bright waveform, one with nearly all harmonics present. Sawtooth and narrow rectangular waveforms are good. If you have a modular synthesizer with more than one oscillator, or if the oscillator has more than one waveform output, try feeding the oscillator outputs through separate filters or modifiers and mixing the filtered outputs. Your overall equalization should include some mild high-pass filtering to attenuate the lowest octave of fundamental, and some mild low-pass filtering above 8kHz or so.
Animation (frequency and waveform modulation) is important in string tone simulation, but must be employed with care. Vibrato must be continuously shaped by increasing both rate and intensity as each individual note progresses, and by withholding vibrato entirely when fast passages are being played. Delayed vibrato that builds up from zero frequency modulation every time a new note is depressed is also useful. If you have a modular synthesizer with voltage-controllable rectangular waveform width, try modulating the waveform very slowly to simulate the sorts of variations a violin player achieves by bowing different points on the string.
String instrument envelopes are generally simple—slow rise, full sustain level, and medium decay. It is generally only necessary to shape the overall loudness, although you may want to employ a slight bit of low-pass filter envelope to make the sound get brighter as it gets louder. Of course, vigorous, rapid violin tones do not have slow attacks. The starting transients of such tones are brief and have a high noise content. If you have a modular synthesizer or are experimentally inclined, try shaping pink noise with a very short envelope and then applying the shaped noise to modulate the oscillator's frequency. This will give a "gritty" attack transient that suggests vigorous bowing.
Simulating string instrument tones is a complex business that requires a lot of experimentation. Although filtering requires special attention, no one aspect of string tone simulation is so important that its proper use alone is enough to achieve good results. The interactions among waveform, animation, enveloping, filtering, and of course appropriate phrasing must be developed with taste and care. You may never actually wind up with a really good string tone simulation, but you will learn a lot about sound while you work.
My next column will get into another complex area: vocal tones.
For more articles by Bob Moog, please visit http://www.keyboardmag.com/analog/1318/moog-monday-hub-page/57761