A realtime .vst synthesizer must process input MIDI and return an audio buffer in < 1ms. This meant that, when designing grainferno, I had to be extremely careful about even the tiniest of details. Stack frame creation, conditionals, memory allocation, and complex operations needed to be avoided or optimized. I used profiling tools to eliminate bottlenecks, and used precomputed LUT buffers, lazy modulator computation, and other techniques to ensure glitch-free audio and low CPU usage.

Much of the DSP is based on sinc-interpolation: this is what allows grains to be dynamically repitched, and also what allows them to "start" in-between sample positions. However, there were many other more minor DSP consideration, such as Biquad Filtering, Resampling (for audio files which didn't match the host sample rate), and RMS calculation for visualization

Modern synths often offer flexible, drag and drop modulations, and grainferno is no exception. Various modulators have been built-in to grainferno: Enevelopes, Drawable LFOs, Randomness & More, have been implemented from scratch, & can dynamically be linked to any audio parameter. Such flexibility means that DSP computations cannot work with static (per buffer) parameter values: modulations must first be applied, from various sources. Additionally, the state of these modulators may differ per synth voice. JUCE Parameters are held in a Modulation Matrix class which can load up voice-specific state, and then return the modulated parameter values at a specific point in time.

From the very beginning, grainferno was designed with the end-user in mind. This meant that it needed to be usable even for users with no programming experience, and limited systems and synthesizer knowlege. For every beta, I've made sure the .vst3 or .au file would automatically pass validations and not trigger any serious anti-malware. The most recent betas have also featured MacOS and Windows installers. I've also made various release and demo videos, explaining how to install and use the software. All of this has been organized via discord, where users (over 400) can also provide feedback regarding bugs or feature suggestions. I plan to release the plugin as a paid product when development has finished. The page for that is already in the works!

I designed the GUI from scratch to be unique and interactive, while still being intuitive. Grain generation feedback is shown over the sample window (grain position, amplitude, and compression) shown above, and per grain RMS content is shown behind the envelopes. Modulation amounts (live and range) are also displayed, and LFOs are dynamically drawable.

I wanted grainferno to provide high quality sound and an intuivitve interface, but also some brand new features for users to play with. Initially this was "tonal granulation": the ability to generate grains at audio rate, keytracked to the note being played. This required grain positions to be able to exist between sample positions (solved with sinc interpolation/fractional delays). However, months into development, Serum 2 brought this exact feature to users with an even more powerful modulation system, and additional FX. To remain competitive, I developed "Granular Compression", a novel granular feature, which compresses individual grain upwards in volume.