It lacks hanning windowing and 4x overlaps
It's CPU hungry because it uses 256 instrument voices polyphony.
For me it uses around 73% CPU.
Having more partials than 256 sinewaves would kill the CPU.
The (Frequency width) knob adjusts how wide the individual frequency bands are before using (Freak Detector) on them.
FFT analysis is monophonic.
Sinewave additive resynthesis is polyphonic.
And the (Voice Info) module is required along with some simple math like scaling to create data for the (FFT filter) which is polyphonic like a filterbank.
The (Frequency width) knob directly adjusts the (FFT filter) settings.
Pitch tracking + volume envelope tracking is done by using the (Freak Detector) from SalamanderAnagram?
FFT analysis done by using (Ez FFT 512).
Phases are discarded for additive resynthesis.
It's not a phase vocoder.
It's more towards sounding like sinusoids.
It has spectral leakage + that modulator sound because i don't how to create hanning windowing and 4x overlaps on the primary sinewave oscillator from scratch.
I don't know how to stop those fake partials from appearing where they was not present in the original audio file. Probably due to spectral leakage?
Reaktor should be able to sound better than (Klingbeil SPEAR), because REAKTOR features core cells.