DSP written for the audio callback.
This is the core of the practice. Filters, dynamics, pitch and time, spatial processing, synthesis, and analysis, written to run inside the audio callback without allocating, locking, or drifting.
The engine behind it is a C++ codebase with 867 passing tests, and a pitch and time engine that moves pitch and transients independently at roughly 9.8 times real time.
Equalisation, compression, saturation, reverb, delay, modulation, pitch correction, time stretching, and synthesis. Built new, or drawn from the existing library and shaped to your product.
Onset and pitch detection, loudness and true peak, spectral analysis, and the measurement rigs that tell you whether a change actually improved anything.
No allocation, no locks, and no unbounded work on the audio thread. Denormals handled, state properly reset, and parameter changes smoothed so automation does not produce clicks.
Null tests against a reference, offline regression on a fixed corpus, and listening tests in the DAFx style where the question is perceptual rather than numerical.
Yes. Several parts of the engine are available as source under a licence, including the pitch and formant work.
For most processing, one buffer. Where an algorithm needs lookahead we tell you exactly how much and why.
Often. We implement from the literature and test against the published results where the paper reports any.