Historically, only humans were capable of creative action to the extent of composing music. However, machine learning is challenging prior paradigms by leading computers to the frontier of human-level intelligence. Computer generation of music began in 1951, when Alan Turing transposed music for a computer to perform (National Computing Laboratory, Manchester, 1951). Considerable progress was made only recently due to the development of the graphical processing unit and deep learning in the early 2000s. Most efforts have emphasized melodic and harmonic style (CITE), but there has been little progress in terms of thematic variation within a style. By utilizing Google's TensorFlow library, Magenta, I trained a single, deep recurrent network to model strongly thematic melodies, using a diverse corpus of video game music. This challenged the capabilities of recurrent neural networks since throughout a game many themes are presented, ranging from war, defeat, and victory, to love and death. The aesthetics of music are ultimately subjective; nevertheless, my analysis, in an attempt to mitigate bias, evaluated generated songs on modulation, strength of rhythm, and repetition. This deep generative model, trained on 2,933 video game melodies, recognized unique themes, and composed compelling and distinct themes.

git clone https://github.com/tensorflow/magenta.git follow install procedure install bazel

TO GENERATE NOTE SEQUENCES::: ./bazel-bin/magenta/scripts/convert_dir_to_note_sequences --input_dir=~/ee/midi/ --output_file=notesequences.tfrecord --recursive

BACKGROUND::: 1. CTRL Z 2. bg 1 3. disown -h %1

0. cd magenta/
1. ./bazel-bin/magenta/models/melody_rnn/melody_rnn_create_dataset --config=attention_rnn --input=/home/spencer_l_churchill/ee/notesequences.tfrecord --output_dir=/home/spencer_l_churchill/ee/out/ --eval_ratio=0.10
2. ./bazel-bin/magenta/models/melody_rnn/melody_rnn_train --config=attention_rnn --run_dir=/home/spencer_l_churchill/ee/rundir/ --sequence_example_file=/home/spencer_l_churchill/ee/out/training_melodies.tfrecord --hparams="batch_size=64,rnn_layer_sizes=[13,64,64,13]"
3. tensorboard --port 6969 --logdir=/home/spencer_l_churchill/ee/rundir/
4. ./bazel-bin/magenta/models/melody_rnn/melody_rnn_generate --config=attention_rnn --run_dir=/home/spencer_l_churchill/ee/rundir/ --output_dir=/home/spencer_l_churchill/ee/generated/ --num_outputs=10 --num_steps=480 --hparams="batch_size=64,rnn_layer_sizes=[13,64,64,13]" --primer_melody="[]"

0. cd magenta/
1. ./bazel-bin/magenta/models/polyphony_rnn/polyphony_rnn_create_dataset --input=/home/spencer_l_churchill/ee/res/notesequences.tfrecord --output_dir=/home/spencer_l_churchill/ee/res/out/ --eval_ratio=0.10
2. ./bazel-bin/magenta/models/polyphony_rnn/polyphony_rnn_train --run_dir=/home/spencer_l_churchill/ee/res/rundir/ --sequence_example_file=/home/spencer_l_churchill/ee/res/out/training_poly_tracks.tfrecord --hparams="batch_size=64,rnn_layer_sizes=[24,13,13,24]"
3. tensorboard --port 6969 --logdir=/home/spencer_l_churchill/ee/res/rundir/
4. ./bazel-bin/magenta/models/polyphony_rnn/polyphony_rnn_generate --run_dir=/home/spencer_l_churchill/ee/res/rundir/ --output_dir=/home/spencer_l_churchill/ee/res/generated/ --num_outputs=10 --num_steps=128 --hparams="batch_size=64,rnn_layer_sizes=[24,13,13,24]" --primer_melody="[]"