Fuzzy MIDI

This year's Winter NAMM marked the 20th anniversary of MIDI, which was first demonstrated at the 1983 show when a Roland JX-3P and Sequential Circuits

This year's Winter NAMM marked the 20th anniversary of MIDI, which was first demonstrated at the 1983 show when a Roland JX-3P and Sequential Circuits Prophet 600 were connected with a MIDI cable and played as one instrument. Since then, many new features and capabilities have been added to the MIDI Specification, such as MIDI Time Code and MIDI Machine Control, clearly demonstrating the expandability and robust nature of this now-ubiquitous interface.

During the show each January, the MIDI Manufacturers Association (MMA) meets to consider proposals for new Control Change, Universal System Exclusive, and other messages. This year, there was one proposal that particularly caught my attention. It was submitted by General Mai Dai, a former officer in the Vietnamese army and now president of a music-technology and military think tank called M1D1 (“Music First, Defense also First”). He called his proposal Fuzzy MIDI.

General Dai started by explaining the concepts of fuzzy logic and neural networks, on which his proposal is based. The basic premise is to transcend the 1/0, yes/no left-brain binary orientation of conventional computing and instead process information in a parallel right-brain manner, using variables with many possible values. In addition, several variables can be considered simultaneously, each weighted differently to affect the outcome of the operation.

Dai's proposal applies these concepts to MIDI, with many new messages. For example, a new Control Change message called Note Percent would indicate the percentage by which the note is on or off. As a key is played on a keyboard, its entire movement from the Off point to the On point is tracked and used to control any number of parameters. In addition, the proposal defines a new System Real Time message: Rubato. This message will allow fuzzy sequencers of the future to follow your lead as you play with widely varying (or entirely absent) tempos.

You could argue that MIDI is already a bit fuzzy. Most of the variables in MIDI can take 1 of 128 different values, and some many more than that, in order to simulate analog continuity. But it doesn't always work very well, and most variables are limited to 7-bit resolution. In response to this, Dai's proposal suggests that an analog control-voltage representation of MIDI messages be created for true fuzziness. This provides infinite resolution, and it can even be transmitted on the two unused wires in a standard MIDI cable (see Fig. 1).

Also included in the proposal are the preliminary design specs of a fuzzy MIDI-mapping device in which several variables are considered before the mapping is established. For example, you could program it to accept Mod Wheel, Aftertouch, and Note Percent messages and define the relative strength of their effect on a mapping to Pitch Bend, Breath Controller, and Volume. Such a device has obvious applications in algorithmic composition as well as live performance and recording.

Interestingly, the furor that followed Dai's presentation was phenomenal. There were many fuzzy thinkers in the group who defended the proposal at the top of their lungs, while those opposed to the idea accused its supporters of being neurotic about neural networks.

The ruckus was still raging as I stole away and returned to my room at the Weisenheimer Inn, where I could peruse the proposal without having to avoid flying bran muffins and juice pitchers. Later, General Mai Dai stopped by, so we went to the bar and talked into the wee hours over more than one mai tai. By morning, my head was a little fuzzy, but I was convinced that his proposal was worth pursuing, and I look forward to seeing it implemented in every MIDI device made from now on.