U-HE has released DIVA - http://www.u-he.com/cms/113-diva. The have also extended the discount to Jan. 15th. I just bought it and it sounds fantastic!!!
The oscillators, filters and envelopes closely model components found in some of the great monophonic and polyphonic synthesizers of yesteryear. Modules can be mixed and matched so you can build hybrids, but what sets DIVA apart is the sheer authenticity of the analogue sound. This comes at the cost of quite a high CPU-hit, but we think it was worth it: Diva is the first native software synth that applies methods from industrial circuit simulators (e.g. PSpice) in realtime. The behaviour of zero-delay-feedback filters when pushed to the limit clearly demonstrates the advantages of this groundbreaking approach.
Diva works without restriction until December 31st 2011. From then on you will notice a mild crackle, the demo restriction that will remind you to visit our shop.
Normal price: 179$ (+19% VAT in EU)
Introductory price: 119$ (+19% VAT in EU) extended till January 15th, 2012
About Zero Delay Feedback Filters
I found the following info in the quickstart.
Most filters in old analogue synthesizers have one or more feedback paths. The input signal passes through a series of controlled resistors and condensers, and the output is fed back into the input (or to another point within the signal path). The classic Moog ladder filter has four simple lowpass filter stages and a resonance control that determines how much of the signal is fed back into the input. This feedback loop doesn’t introduce any delay, feedback is practically immediate...
Digital models try to reproduce this by calculating the result of applying four simple lowpass filters to an input sample. Feedback means repeating the calculation using the results of the previous one.
While real circuitry can process a signal within a few nanoseconds, digital models calculate per sample, adding in the region of a million times as much latency to each iteration. The cumulative effect of feedback latency in digital emulations is very noticeable. Oversampling and higher sample rates help reduce the latency, but conventional digital filters always “smear” at higher resonance levels.
Ways to address the latency problem have been available for many years, but implementations either don’t model the inherent complex distortion inherent in real circuitry, or they aren’t suitable for realtime processing. Common to all methods is that they predict output values and use that prediction for the current calculation
Of course we can’t see into the future, but our routines are fast enough to deliver at least a few voices in realtime. We calculate the filter with a few test samples and look at the deviation between the prediction and the result. We use that deviation to calculate a better prediction – the goal is to close the gap as quickly as possible!
We have applied a classic “trial and error” principle with a (we like to think) rather intelligent method of learning from our “mistakes”. As a rule, the filters in Diva have to be calculated 3 to 4 times in succession - but it can take up to 20 cycles e.g. at high resonance or if the inputincludes white/pink noise.
Mindmap of DIVA Anatomy
I created a mindmap showing the anatomy of the oscillators, filters and envelopes in relationship to the model template. View the map full screen by clicking here or use the viewer below.