Mirroring

An orchestra of captured moments made from the sounds of Detroit.

Mirroring: An Orchestra of

Captured Moments

Every audio sample in the Mirroring concerto will be gathered from found sounds throughout the city of Detroit. I am training an AI from scratch on all of the music I have ever written. Then, I will be iteratively composing with an amalgamation of all the previous versions of my musical writings to create a concerto between the sounds of the city and a live (human) soloist. This work will premiere in the spring of 2026! If you have a sound you would like to contribute in the city of Detroit, please don’t hesitate to reach out under the contact page.

1: Model’s MIDI output (notes, velocity, duration, inter-onset-interval IOI across 10 instruments)
2: Me curating the sound samples used on the exact same MIDI multitrack (the notes, velocity, duration, IOI remain exactly the model’s output)
3: My compositional iteration

After testing many architecture approaches including LSTM/BiLSTM/Attention Layers/GPT-2 Transformers, I have settled on a Temporal Convolution Network (TCN) for this project. Trained on a small data set (which I am expanding) of my own compositions, I (1) take the model output as a song seed. From this I (2) grow an arrangement of lush samples and (3) iterate to produce a compositional form.

For full blog of how we got here visit: https://summerkrinsky.com/computermom