More Ensemble Stuff (long reply)

[Sean Costello]

Paul Perry wrote:
> 
> I wonder whether regular LFOs are the best?
> Might some very low frequency noise source do better?
> If anyone can check this in c-sound or otherwise, I'd like
> to know. Intuition suggests that 'fractal' (one/F) noise
> would be best.
> "regular' fixed freq LFOs are the one thing that I HATE about
> most synths & electronic music.
> 
> paul perry melbourne australia

For a lot of applications, I think that low frequency noise is the best
modulation source.  It sounds great, for example, for slowly modulating the
delay lengths in feedback delay network reverberator structures. However, I
don't think it would work very well as the sole modulation source for
ensemble-type effects, at least if you are trying to simulate a string orchestra
or vocal chorus. A real string ensemble or vocal choir will have several
different types of pitch modulation for each source:

- General tuning discrepancies between each source. No one is going to sing or
play the exact same note. This is similar to a multi-VCO synth, with each
oscillator tuned to a slightly different frequency.
- Tuning drift and jitter. Each instrument or voice will have a tendancy to
drift slowly as time goes by. In addition, the human voice has some higher
frequency "jitter" that is the result of twitching muscles in the vocal tract.
The drift and jitter are similar to that seen with real-world VCOs.
- Vibrato. Each instrument in a string ensemble or vocal choir will almost
inevitably have a certain amount of vibrato. This is a modulating signal,
between 4 and 7 Hz, that is fairly close to sinusoidal in nature. Each
instrument will have a slightly different vibrato rate, and the vibratos will be
out of phase with each other.

As far as modulating delay lines to get an "ensemble" effect, the general tuning
discrepancies and tuning drift can be modelled quite well with either a
multiphase LFO of low frequency (1 Hz or lower), several LFOs (1 for each delay
line), or several uncorrelated low-frequency noise generators.  Using 1/f noise
will result in an extremely noisy signal, instead of a gentle undulation of
pitch. Two ways of generating a useful low-frequency modulation signal:

1) White noise is sent through a lowpass filter with a very low cutoff (below 10
Hz).
2) White noise is sent through a sample-and-hold unit, with the sampling
controlled by a very low frequency clock. The output of the S/H is sent through
a lag unit, to form a waveform that slowly "connects" the dots between points of
known frequency, but random value. This is how randi works in Csound.

However, trying to use a random low-frequency generator to create the "vibrato"
component of the signal just doesn't work. A randi generator, set at a rate of 6
Hz or so, does not sound like a natural vibrato - it sounds creepy, and seasick,
and all sorts of things that have their time and place, but just don't work in
an ensemble sound. I have tried using randi generators in my ensemble circuits
to create the vibrato sounds, and the result just sounds "off." It is
interesting that the ear can pick up on this, considering how complex the sound
is. Using several randi generators makes a sound that is definitely electronic,
not particularly musical, and is definitely unsettling - somewhere in between a
broken tape deck and a seasick orchestra.

On the other hand, white noise filtered through a bandpass filter, with center
frequency of a desired vibrato rate and very high Q, might do the trick.
Apparently this results in a signal that sounds more vibrato-like than the randi
generators, or lowpass filtered noise. Perhaps using 1 of these per delay line
would result in a nice sound.

Another thing to remember about the ensemble effect is that the multiple LFOs,
or multiphase LFOs, really do a nice job at concealing the periodicity of the
individual LFOs. Once you get beyond 2 LFOs, it is really hard to pick up the
pattern between the individual LFOs. The two 3-phase LFOs set at different
frequencies really creates a complex sound, especially if the modulation amounts
are set right.

Probably the ideal modulation scenario would be to use 4 modulation sources:
- 3 individual random LFOs, set for the slow range of modulation (below 1 Hz).
These would contribute to the general drift and tuning discrepancies of the
ensemble.
- 1 multiphase LFO, set to somewhere between 4 and 7 Hz, depending on taste (I
prefer slower vibratos, for a more Gary Numan-esque sound). This would provide
the essential vibrato component.

A completely random alternative would use either 3 bandpass-filtered noise
units, or 3 very low frequency random LFOs combined with 3 bandpass-filtered
noise units for the vibrato. Of course, this adds a great deal to the complexity
of the circuit, compared to 6 op-amp stages for three independent LFOs, or a
single CD4069 for the two 3-phase LFOs.

Sean Costello