This is early results of exploring the highly entropic chorales created by the TonicNet model.
I built a chorale generating notebook in python that created 4900 examples using the TonicNet model, and then ran those through an evaluation routine (using muspy) to find those that had the highest degree of pitch entropy. There were at least 1000 that included all 12 tones in the tempered scale.
I then chose a few that were in the key of D major. All were strange and wonderful chorales. TonicNet writes them out as MIDI files, with a kind of piano-roll format of four voices and a certain number of notes in each voice, all 1/16th notes. If a note is being played, then a MIDI number appears in the slot for that time-step.
You need some logic to turn this piano roll type notation into notes with duration.
I then repeat each note 15 times, turning every 1/16th note into 1/16th less than a whole note. I then apply masks to turn notes off to create arpeggiations. Or for the woodwinds, I just have long held notes.
I transposed the Victorial Rational Well Temperament from the scala scale archive into the key of D. Some of the ratios may seem kind of extreme, but that’s what was required to accurately reflect the ratios in the temperament when transposed. This is the result of that:
George Secor's Victorian rational well-temperament (based on Ellis #2) in D
I then created a finger piano arpeggio vamp with eight voices, and added a double woodwind quartet (oboes, clarinets, french horns, bassoons) playing slow chords. Both voices simply took the notes that the TonicNet model created. I modified some characteristics, including envelopes, volume, timbre, and other factors. The features are changed at the 1/3 and 2/3 points in the piece. The result is a sweet sounding exploration of what the model thought Bach might do.