[Release] DephazEAudi0 — real-time geometric audio processor

[jamcat]

So this is a smoothing algorithm? Some kind of “analogizer”?

Provide an image of a waveform before and after processing.

[dephazeaudio]

Not really — it’s neither a smoothing stage nor an “analogizer” in the traditional sense.

There’s no temporal averaging, filtering, or attempt to emulate analog behavior. The processor applies a deterministic nonlinear mapping to each sample independently, so the effect is structural rather than a time-based smoothing process.
Waveform comparisons are definitely possible — and honestly, the most meaningful results come from testing with your own material.

The visual differences aren’t fixed or universal because they depend heavily on the source signal. On some material they’re subtle, on others they’re quite obvious.

So rather than presenting a single “proof” image, it makes more sense to evaluate it in context — both by listening and by analyzing signals you’re familiar with.

[jakke]

https://github.com/angusdewer/DephazEAu ... zEAudiO.py

The executable provided in the same repository is very similar to the plugin in the Gumroad video. The gain control works basically like a normal gain with compression through a process wherein the audio sample values are mapped into vectors in different "spatial curvatures" (like stretching a unit circle into a unit ellipse, for example) and the difference of their length is then used as a gain factor, somewhat simplifiedly. The stereo spread is exactly like any other basic M/S stereo widener.

Take that as you will.

[Dr.Gunjah]

I apologize you had to experience kvr default state...juvenile trolling...rational behavior and cordial objective discourse can be hard to come by...there is some modicum of value if you put on ur rubber boots and wade through the sewage...

Kinda hard to have an objective discourse on this plugin when the dev can't give clear answers, examples or really any useful information about it. Everything is covered in blurb and 'explained' with esoteric descriptions.
I'm with Tj here, this looks like some sort of psychological test.

[oobesan]

If “Gain” doesn’t add gain to the signal, maybe call it something not “Gain”.

Part of the job of selling something is being able to communicate what the thing you’re selling is going to do for your potential customers. I think we can all understand waveform transform functions. However, I don’t anyone is sure what “improves coherence” means.

[dephazeaudio]

Thanks for taking the time to look at the Python reference code — I appreciate the effort.

However, your interpretation of what the mapping is doing isn’t quite accurate. The transform is not behaving like a simple compressed gain stage or ellipse stretching model. While it may resemble that superficially when reduced to scalar math, the actual mapping is structured differently and was intentionally designed around a geometric transform model rather than conventional gain shaping.

The Python code is published under my copyright precisely so people can inspect and experiment with it, but interpreting nonlinear transforms can be tricky — different simplifications can lead to misleading conclusions.

If you’re interested in discussing the math in detail, I’m happy to clarify specific parts of the algorithm.



Technically, gain just means an amplification factor — a numeric scaling applied to a signal. In most audio tools that scaling directly affects loudness, which is why people naturally expect a “Gain” control to change volume.

In this plugin, the control adjusts the intensity of the internal transform rather than acting as a simple output level. So mathematically it’s still a gain factor, but perceptually it behaves differently than a traditional volume control.

I understand how that naming can create an expectation mismatch, and clearer labeling is something I’m considering based on feedback like this.