Cross Adaptive Audio Versions Save

Evolving Artificial Neural Networks for Cross-Adaptive Audio Effects

0.7

7 years ago
  • Box-and-whisker plot
  • Fix effect visualization bugs
  • Mono output from live csd
  • matplotlib.use('Agg') for certain Linux environments

0.6

7 years ago
  • Plot aggregated fitness values in a 2D grid (heat map), visualizing the result of a grid search (hyperparameter optimization)
  • Fix live mode for non-composite effects
  • Visualize composite effect mix values
  • Visualize effects and their parameters. Animate knob values while playing the sound.
  • Disable species visualization when too much data
  • Add a utility for scoring testing data
  • Fix some bugs

0.5

7 years ago
  • Apply a set of effect in parallel and/or serial. Effect templates were refactored to make this possible.
  • Add Csound analyzer which is also able to run in live mode (for performances) as well as offline (for training)
  • Various k-rate variable interpolation fixes
  • Visualize species in web client
  • Add more command line arguments, such as --keep-csd, --verbose, --crossover-rate and --mutation-rate
  • Add a set of effects, such as flanger, reverb, various distortion effects and ring modulation
  • Add support for duplicate parameter names in web client
  • Add more test sounds, especially short ones, which take less time to process
  • Tweak some default GA parameters (higher probability of weight mutation and structural mutation)
  • New neural mode allows you to evolve targets for each time step with only static input
  • Add experimental live mode which is compatible with CsoundAnalyzer and a very limited set of parameters
  • Change colors in fitness plotter
  • Add a data augmentation command for augmenting short sounds

0.4.1

7 years ago
  • Show more info on experiment card
  • Update readme
  • More audio effects
  • More experiment settings
  • Add mixed fitness mode
  • Tweak plots
  • Improve performance
  • Add try_effect.py

0.4

7 years ago
  • Implement novelty search
  • Support for storing and visualizing multiple experiments
  • Check sanity of input files
  • Calculate and visualize similarity objective regardless of chosen fitness function

0.3.2

7 years ago
  • Introduce weights in similarity config. This means you can make for example errors in amplitude count more
  • Deal with completely silent output files correctly

0.3.1

8 years ago
  • Improve readme
  • Clean up
  • Minor fixes

0.3

8 years ago
  • Multi-Objective fitness evaluation inspired by NSGA-II
  • A command for listing all analyzers and features
  • Correct effect parameter mapping
  • Misc bug fixes and improvements

0.2.1

8 years ago
  • Add support for essentia and sonic annotator with libXtract as the main plugin
  • Remove the separate make project command. It is now integrated into the neuroevolution program.

0.2

8 years ago
  • A web client interactively visualizes results
  • Parallel computing and other performance fixes (~4x speedup)
  • Some math that was broken in 0.1 now works correctly
  • Neural input can easily be tuned
  • Adding new effects is easy now