Diva Dataset

This repository contains data and scripts for exporting, analyzing the presets of Diva synthesizer.

Explanation of Files and Directories

Data

dataset/dataset.json: This file contains all the information about preset parameters and labels. It comes from dataset.json in flow synthesizer repository.

Executable python scripts:

export_dataset.py: Exports the dataset to diva-preset-audio-dataset/ directory. You can also uncomment dump_features() to dump audio features directly to dataset/features.pkl.
plot_features.py: Plots the features extracted by dump_features() function.
extract_mel.py: Uses the data in the diva-preset-audio-dataset/ directory, select all the presets that have at least one timbre label, extracts the Mel-spectrogram and the labels, dumps dataset/mel.pkl and dataset/labels.pkl.
analyze_dataset: Analyzes the labels dumped as dataset/labels.pkl.
model.py: Defines several machine learning models and uses dataset/mel.pkl and dataset/labels.pkl as training data. Use python model.py <model name> to run the script. The model will be put into models/ directory.
analyze_latent.py: Analyzes the latent space of the auto-encoder model.

Utility script:

dataset_parser.py: Provides utility functions to parse dataset/dataset.json and convert the label format.
config.py: Some constant definitions.

Other scripts

osc_handler.py: Handles the OSC message communication with Ideator plugin host.
preset_parser.py: Dedicates to parse the original preset format used by Diva synthesizer. The function is not complete for now.

To export the audio, you have to compile RenderMan.

Configuration is for MacOS:

Install boost libs: brew install boost-python3 (My current version is 1.73.0)
Install JUCE (the version should be JUCE 5)
Make sure have python3.8 installed
Get RenderMan from https://github.com/fedden/RenderMan
Open RenderMan-py36.jucer in the project folder, generate XCode project.
Configure the search paths in XCode (This could be different, find the corresponding paths on your machine):
- Boost dylib path: /usr/local/Cellar/boost-python3/1.73.0/lib
- Python dylib path: /usr/local/Cellar/python@3.8/3.8.5/Frameworks/Python.framework/Versions/3.8/lib
- Python header path: /usr/local/Cellar/python@3.8/3.8.5/Frameworks/Python.framework/Versions/3.8/include/python3.8
Set the link flags: -shared -lboost_python38 -undefined dynamic_lookup
Build a release build librenderman.so.dylib, remove the extra .dylib extension.
Copy the librenderman.so to your python project root directory.