Generative Model Samples
This is a collection of samples created by generative machine learning models.
Hopefully it can be useful for anyone who quickly wants to compare the output of some models without having to dig through a bunch of research papers, code, and GitHub repos.
This is not a comprehensive collection, and will not aim to be. The goal is to provide a "best effort" snapshot of existing generative modelling capabilities, using the author's checkpoints when possible. The repo will be updated as newer and better content is produced, so pull requests are appreciated.
In the case of closed source models, or models with no pre-trained checkpoint, samples are those publicly released by the developers, and are provided as-is. For open-source models, samples were generated by running on a local machine, with commands provided in the sh
directory.
Some models can take a primer, which acts as prior knowledge when performing generation. Folders for these models will denote no-primer
or primer
if that was tested.
If a pre-trained model was available, it was used to do the generation. In some cases, several model configurations are available. The commands in sh
can be referenced for specifics about which configurations were used.
Coverage
Model | Domain | Output Format | Open Source? | Pre-trained? | Runtime Commit ID | Generation Date | Comments |
---|---|---|---|---|---|---|---|
BachBot | Audio | MIDI | Y | N | - | - | Source: Soundcloud |
DeepJazz | Audio | MIDI | Y | N | - | - | Source: Soundcloud |
GRUV | Audio | Waveform | Y | N | - | - | Source: Youtube |
Drums RNN | Audio | MIDI | Y | Y | 0b4b73b | Mar. 23 '18 | |
Improv RNN | Audio | MIDI | Y | Y | 0b4b73b | Mar. 23 '18 | |
Melody RNN | Audio | MIDI | Y | Y | 0b4b73b | Mar. 23 '18 | |
Performance RNN | Audio | MIDI | Y | Y | 0b4b73b | Mar. 23 '18 | |
PianoRoll RNN-NADE | Audio | MIDI | Y | Y | 0b4b73b | Mar. 23 '18 | |
Polyphony RNN | Audio | MIDI | Y | Y | 0b4b73b | Mar. 23 '18 | |
Wavenet [ibab] | Audio | Waveform | Y | DiyuanLu | e11ad19 | Mar. 23 '18 | Tried with conditioning (Speaker#300) & without |
Onsets and Frames | Transcription | MIDI | Y | Y | 0b4b73b | Mar. 20 '18 | Piano transcription, .wav to .midi |
Testing
MIDI and Waveform file playback was tested using Qmmp on Ubuntu 16. All files should play correctly.