TurboAlexis

TurboAlexis - MidiVerb1/2 and Midifex Emulator Plugin.

A JUCE-based audio plugin that emulates three classic MidiVerb/MidiVerb2/MidiFex reverb effects. Based on the code by https://github.com/thement/midiverb_emulator, where most of the legwork is done.

It inherits the TurboPaco (my other reverb) philosophy (and UI). Few controls, fun to program and if something doesn't work then change program.

What adds to "midiverb_emulator" from Thement.

• "fancy" UI with many usability improvements. Mousewheel working everywhere, drop-downs arrows, ability to write values, etc.
• No mix knob. Send mode (dry=0%) or Insert mode (dry=100%) with Wet level.
• Clip level knob to do gain-compensated gain staging, as the FX can be 13 or 16-bit only, so it needs to be gain staged properly at 0dBFS.
• Ability to change the size of each algorithm/program from 25% to 200%. This expands the usability of many algorithms. e.g. echoes, non-linear reverbs, etc.
• As the algorithm size is variable, it adds delay line interpolation. Selectable between Zero Order Hold (none), Linear, Cubic Spline, and Thiran.
• Adds reverb modulation with selectable type, rate, and amount. See chapter below.
• Can use float or 16-bit fixed point arithmetic. Both with optional 13 or 16-bit quantization at the FX input. 13 bit quantization can use a noise shaper. The 16-bit one can use a dither (mostly useful either for lofi or the choruses/flangers when deliberately gain-staging wrongly).
• Allows over and undercloking the effect sample rate. This allows both lo-fi and "hi-fi" effects, as the devices running at 2x the sample rate with the size at 200% are equivalent.
• Changes the "defeat" programs from silence to pass-through.
• Adds predelay parameter.
• The resampler has quality settings to allow aliasing in.
• Serial lowpass and hipass filtering before the input with the ability to amplify the highs, so it can be used as a cheapo preemphasis filter.
• Stereo width and panning.
• Ducker.
• Removed more dead code in the algorithms (and the original device).

About the modulation.

Modulating the delay lines is possible-

The approach taken is that I haven't tuned each algorithm/program individually
to find what works best, it's a lot of work.

What is done instead is to provide 8 different combinations of modulation. What
they do is to modulate delay lines based on their index and size. Each of the 8
alternatives has different combinations, so it's up to the user to choose if there
is one they like.

4 interpolation modes are provided: Allpass, Catmull-rom spline, Linear
interpolation (lerp) and Zero Order Hold (ZOH, no interpolation). All of them
sound different.

Settings matching the original hardware.

I have never owned it or use it, but based on what I know:

• Mfx and Mv1: No modulation, ZOH, 13 bit A/D, vintage or lo-fi resampler, maybe a tad of highs over 100% as preemphasis. Very sensitive to gain staging, so the clip level should be set accordingly.
• Mv2: No modulation, ZOH, 16-bit A/D, vintage or lo-fi resampler.

LFO based programs.

All the LFO based programs (choruses, flangeres) don't support the triggered
flanges, as the parent emulation project doesn't support it either.

The LFO amount is configurable using the size knob (it uses the internal
emulated LFOs). The program's original is the parameter default: 100%.

The LFO rate can also be changed. The program's original is the parameter
default: 50%.

There is an added feedback control for the LFO based programs not present
on the original device.

Sample rates.

It's not an exact match. The Mv1 and Mfx use 23.4Hz (vs ~23.44 on the device) and the Mv2 uses 31.5kHz (vs 31.25 on the device).

The reason is simple, many fractional resamplers have a fixed set of shifted windowed sinc tables at different offsets, (e.g. 256 tables) and use interpolation to get the in-between values. Those tables have non-contiguous (but predictable) memory access patterns.

The resampler I use instead calculates the period of the resampling cycle and uses a single table for each stage in ascending order. This gets rid of the need for interpolation, allows an ascending prefetcher-friendly access pattern and can have most of the time less tables than the conventional approach for common samplerates.

It has the drawback that for very "unfriendly-to-48-or-44k" rates there are more tables generated. The frequencies chosen to run the device at where chosen for being near enough and require a very low amout of tables.

Know something about the original hardware or want to contribute something?

Contact me. All information is welcome. E.g. about the triggered flanges, MV2 input/output filters, etc.