The basic idea of Amalgam is using two long delay lines (created by stringing two and three delay lines in series for each side) as a kind of out of phase “tape looping” platform. As audio is entered into both delay chains, it begins to phase, and the relationship between sounds on either side can become permutations of the other, as the length of the loop changes new sounds’ temporal relationship to sounds already held in the loop buffer. And every so often, the micro-composition formed from the accumulation of sounds over time fades away, and a new one begins.
The sound engine is a three-voice triangle-wave “chord” generator. (The chords are generated by quantizers run in series, with a small random value being applied to the input of each successive quantizer. The quantizers are tuned to pentatonic scales, to give a better shot at producing pleasing chords from the fairly haphazard generation method.) Each triangle-wave oscillator is feeding back on itself, controlled by a slow sine wave LFO. Then they pass through a 12 dB/oct low-pass filter, also modulated by the envelope as well as… a slow triangle wave LFO!
The output of this chord voice is then fed into a VCA, controlled by the same envelope that affects the filter. The output of the VCA is connected to both sides of the out-of-phase delay looping contraption, but the envelope _also_ ducks the VCA controlling the feedback of the loop, so new sounds can suppress and create odd juxtapositions against sounds already in the loop (the envelope cycles between four different lengths, randomly, and its shape is continually modulated, so a lot of the sounds have a fairly gentle attack and decay, but some are sharper decay envelopes and others are pseudo-backwards-sounding attack envelopes, etc. The envelope itself is trigger probabilistically, based on a “clock” that is unrelated to the speeds of the looping delays, which is intentionally done to create more complex spatial varieties in the loops themselves — the different envelope lengths are divisions of this clock’s speed).
The loops themselves have very gentle high- and low-pass filters in the feedback loop, which will cause sounds to slowly decay and mellow. A small amount of noise (three random modules multiplied against one another to give a more Guassian response) is also applied to the VCAs controlling the feedback loop, which introduces some subtle distortion, particularly in the low-end as the loops regenerate. Then the loops pass through more high-pass filtering before entering a hall reverb.
Every twenty minutes or so (on average — sometimes much longer, sometimes much less time), the VCAs controlling the loop feedback go low, and whatever composition has developed in that time fades away to be replaced with another. When this occurs, a couple of other things can also change: The length of the right side loop can change, which changes the phase relationship between the two sides (some of the phase relationships are fairly simple, like 4:5; others are more complex, like numbers I figured out when building the circuit but I can’t recall off-hand what they are and don’t want to do the math again). The key and scale of the quantizers can also change, switching between D minor pentatonic, D major pentatonic, A minor pentatonic and G major pentatonic.
And that’s the patch. It creates, I think, interesting little compositions that evolve for a while but then don’t overstay their welcome in most cases.