MEMLNaut Firmware: VerbFX with NISPS
Overview
The VerbFX firmware offers exploration of a multi-effects reverb processor, using the NISPS approach to ML.
VerbFX
VerbFX is a multi-effects processor built around a Freeverb-based reverb algorithm with a filterbank and delays. 47 sound processing parameters are controlled by the joystick. Six effect routing options can be individually enabled or disabled via the FX Enable screen.
Signal Flow
Input → FilterBank → Delays (Short / Medium / Long) → Reverb → Wet/Dry Mix → Output
The FilterBank can receive feedback from the delay outputs. The delay outputs can also be routed into the reverb input.
Effect Blocks
- FilterBank — 8 bandpass filters in parallel.
- Reverb — Freeverb algorithm (8 low-pass feedback comb filters + 4 allpass filters)
- Short Delay — 512 sample delay buffer
- Medium Delay — 2048 sample delay buffer
- Long Delay — 16384 sample delay buffer
- Delay → Reverb — Routes the delay mix into the reverb input
Inputs
The processor takes a stereo input. Left and right channels are summed to mono before processing.
Outputs
Mono processed sound is output to both channels of the line out and headphone sockets.
NISPS
When you indicate a reward, you are rewarding (negatively or positively) how a particular joystick position is mapped to the effect sound it’s producing. As you indicate preferences for these settings, the NISPS system will start to tune the mappings towards your preference. See the NISPS documentation for further info.
Physical Controls
Rotary Encoder: Navigate between screensMomentary A Up: Scramble neural networkMomentary A Down: Jolt neural networkMomentary B Up: Positive rewardMomentary B Down: Negative rewardJoystick: ExploreJoystick button: Positive reward placement — drag a sound to a different place and release to positively reward it in this positionToggle A Down: Clear replay memoryKnob X: Learning rateKnob Y: Wet/Dry MixKnob Z: Exploration noise (turned off when all the way to the left)Knob G: Volume
Screens
Navigate between screens using the rotary encoder.
RL
Info on the neural network - the outputs, the current error rate, the number of datapoints in replay memory, and the most recent command.
Focus
Selecting one or more items in the focus menu freezes all unfocused neural network outputs at their current value, so that NISPS only changed the focused values. This let’s you do detailed training, e.g. if you just want to adjust the reverb but not other sounds.
FX Enable
Toggle each of the six effect blocks on or off: FilterBnk, Reverb, ShortDly, MedDly, LongDly, Dly→Verb.
Voice Spaces
Choose a voice space by pushing the rotary encoder to focus the selector. Move the encoder to choose a space and then push again to confirm the selection.
Available voice spaces:
- Default
- Resonant
- Soft
- Cathedral
- Shimmer
- Chamber
- Metallic
- Granular
- Diffuse
- Dark
- Bright
- Harmonic
These voice spaces bias the neural network mappings in particular directions.
NN Inputs
Shows the inputs to the neural network.
Messages
Various messages logging interactions with the system.
Save Model
Tap on a slot to save a model and current voice space,
Load Model
Tap on a slot to load a model.
Input Source
Choose from
- the joystick that matches your hardware (3D or 4D)
- machine listening analysis of the line input (allows the system to adapt to different sounds)
- various configurations of MIDI controllers
Help
Shows what the various controls do.
System Info
System information about the MEMLNaut and current firmware.
MIDI
The firmware sends out the first 16 neural network outputs as MIDI controllers, so you can use the MEMLNaut to control your hardware synths.
Files
Code
https://github.com/MusicallyEmbodiedML/MEMLNaut-NISPS
Firmware Download
https://github.com/MusicallyEmbodiedML/MEMLNaut-NISPS/releases/tag/VerbFX_1.1.0
See the firmware guide for installation instructions.