About project - shepherdvovkes/idmlatentspace GitHub Wiki

1. IDM Latent Space - MediaWiki Documentation
   1. Main Page

IDM Latent Space is a machine learning research project focused on analyzing and generating Intelligent Dance Music (IDM) and electronic music through latent space representations of synthesizer presets.

Table of Contents Project Overview Key Features Repository Structure Core Components Preset Analyzers Data Files Dependencies Technical Architecture Analysis Pipeline Parameter Importance Weighting Dimensionality Reduction Usage Guide Basic Analysis Output Files Machine Learning Integration Loading Vectors Genre Classification MIDI CC Integration Research Applications Generative Models Classification Tasks Clustering Analysis Installation Requirements Repository Setup API Reference OsirisPresetAnalyzer Methods Data Structures Analysis Results Contributing Development Setup Code Standards Research Papers Related Work Citing This Work License External Links

The IDM Latent Space project explores the relationship between synthesizer preset parameters and musical characteristics through dimensionality reduction and machine learning techniques. The project specifically focuses on Access Virus and Osiris synthesizer presets, creating reduced-dimensional representations suitable for AI/ML applications.

• SysEx Analysis: Parse and analyze SysEx preset files
• Dimensionality Reduction: Convert 384-parameter presets into 32D, 64D, and 128D spaces
• Genre Classification: Analyze preset suitability for different electronic music genres
• MIDI CC Mapping: Maintain CC controller relationships for DAW integration
• ML-Ready Outputs: Generate numpy arrays and structured data for machine learning

• updated_preset_analyzer.py - Main SysEx Toolkit-based analyzer
• ModernPresetAnalyzer class - Core analysis functionality
• OsirisPresetAnalyzer class - Specialized Osiris analysis

• osiris_preset.syx - User/target preset for analysis
• osiris_all_presets.syx - Factory presets database
• Generated outputs - Analysis results in multiple formats

• SysEx Toolkit - Primary library for SysEx parsing
• NumPy - Numerical computing and array operations
• JSON - Data serialization and storage

The analysis pipeline follows these stages:

1. SysEx Parsing - Decode binary preset data using SysEx Toolkit
2. Parameter Extraction - Convert raw bytes to normalized parameter values
3. Baseline Comparison - Compare user preset against factory baseline
4. Importance Calculation - Weight parameters by musical significance
5. Space Reduction - Create dimensionally-reduced feature vectors
6. Output Generation - Save results in multiple formats

Parameters are weighted based on their musical significance:

Category	Base Weight	Special Conditions
Filter	+2.0	Cutoff: +3.0, Resonance: +2.5
LFO	+1.5	Rate parameters prioritized
Oscillator	+1.2	Wave and semitone parameters
Effects	+1.0	Distortion: +2.0, Spatial: +1.5
Envelope	+0.8	Attack and decay prioritized

The system creates three reduced spaces:

• 32D Space - Minimal representation for fast processing
• 64D Space - Balanced detail/performance ratio
• 128D Space - High-detail representation for complex analysis

<syntaxhighlight lang="python"> from updated_preset_analyzer import OsirisPresetAnalyzer

1. Initialize analyzer

analyzer = OsirisPresetAnalyzer()

1. Analyze presets

results = analyzer.analyze_osiris_presets( user_preset_file="osiris_preset.syx", factory_sysex_file="osiris_all_presets.syx", baseline_name="Init" )

1. Save results

analyzer.save_osiris_analysis(results, "my_analysis") </syntaxhighlight>

Analysis generates several output files:

File	Format	Purpose
*_complete.json	JSON	Full analysis with metadata
*_32d.npy	NumPy	32-dimensional feature vector
*_64d.npy	NumPy	64-dimensional feature vector
*_128d.npy	NumPy	128-dimensional feature vector
*_report.txt	Text	Human-readable analysis report
*_data.csv	CSV	Tabular data for spreadsheet analysis

<syntaxhighlight lang="python"> import numpy as np import json

1. Load feature vector

vector_64d = np.load('osiris_analysis_64d.npy')

1. Load full analysis

with open('osiris_analysis_complete.json', 'r') as f: analysis = json.load(f)

1. Access metadata

cc_mappings = analysis['reduced_spaces']['64d']['cc_mappings'] genre_readiness = analysis['reduced_spaces']['64d']['dubstep_ready'] </syntaxhighlight>

The system evaluates preset suitability for electronic music genres:

• Dubstep Ready: Filter parameters ≥ 3, LFO parameters ≥ 2
• EDM Ready: Filter parameters ≥ 2, Effects parameters ≥ 1
• Filter Dominance: Ratio of filter parameters in top selections

Preserved CC controller mappings enable DAW automation:

<syntaxhighlight lang="python">

1. Access CC mappings

cc_mappings = results['reduced_spaces']['64d']['cc_mappings']

1. Example: {'CC74': 'filter_cutoff', 'CC71': 'filter_resonance'}

1. Use for MIDI automation in DAW

for cc_number, parameter in cc_mappings.items(): print(f"Automate {parameter} via {cc_number}") </syntaxhighlight>

The reduced-dimensional vectors are suitable for:

• Variational Autoencoders (VAE) - Smooth latent space interpolation
• Generative Adversarial Networks (GAN) - Preset generation
• Diffusion Models - High-quality preset synthesis

Feature vectors enable classification of:

• Genre Compatibility - Electronic music subgenre matching
• Timbral Characteristics - Bright/dark, warm/cold, aggressive/smooth
• Musical Context - Lead, bass, pad, percussion roles

Dimensional reduction enables:

• Preset Families - Group similar-sounding presets
• Parameter Relationships - Identify correlated controls
• Evolution Tracking - Analyze preset development over time

<syntaxhighlight lang="bash"> pip install sysex-toolkit numpy </syntaxhighlight>

<syntaxhighlight lang="bash"> git clone https://github.com/shepherdvovkes/idmlatentspace.git cd idmlatentspace python updated_preset_analyzer.py </syntaxhighlight>

analyze_osiris_presets(user_file, factory_file, baseline_name)

Main analysis function

Parameters:

• user_file (str): Path to user preset SysEx file
• factory_file (str): Path to factory presets SysEx file
• baseline_name (str): Name of baseline preset to compare against

Returns: Analysis results dictionary

compare_presets(user_preset, baseline_preset, threshold)

Compare two presets and identify differences

Parameters:

• user_preset (dict): User preset data
• baseline_preset (dict): Baseline preset data
• threshold (float): Minimum difference threshold (default: 0.01)

Returns: List of significant differences

create_reduced_spaces(significant_changes, target_dimensions)

Generate dimensionally-reduced feature spaces

Parameters:

• significant_changes (list): Parameter differences
• target_dimensions (list): Target dimensions [32,]

Returns: Dictionary of reduced spaces

save_osiris_analysis(results, output_prefix)

Save analysis results in multiple formats

Parameters:

• results (dict): Analysis results
• output_prefix (str): Output filename prefix

Returns: Dictionary of created file paths

<syntaxhighlight lang="json"> { "user_preset": { "metadata": {"preset_name": "User Preset", ...}, "parameters": {"filter_cutoff": {"normalized_value": 0.75, ...}, ...} }, "baseline_preset": { "metadata": {"preset_name": "Init", ...}, "parameters": {...} }, "differences": [ <pre>], "reduced_spaces": { "32d": { "dimension": 32, "feature_vector": [0.85,], "feature_names": ["filter_cutoff",], "cc_mappings": {"CC74": "filter_cutoff", ...}, "dubstep_ready": true, "edm_ready": true } } } </syntaxhighlight>

1. Fork the repository
2. Create feature branch
3. Implement changes with tests
4. Submit pull request

• Python 3.7+ compatibility
• PEP 8 style guidelines
• Type hints for function signatures
• Docstrings for all public methods
• Unit tests for core functionality

• Latent Space Models for Musical Audio - Analysis of dimensional reduction in audio
• Synthesizer Preset Generation using VAEs - Generative modeling approaches
• MIDI Controller Mapping for Electronic Music - CC parameter relationships

<syntaxhighlight lang="bibtex"> @software{idmlatentspace2025, title={IDM Latent Space: Synthesizer Preset Analysis for Machine Learning}, author={Shepherd Vovkes}, year={2025}, url={https://github.com/shepherdvovkes/idmlatentspace}, note={Electronic music analysis toolkit} } </syntaxhighlight>

This project is released under the MIT License. See LICENSE file for details.

• GitHub Repository
• Access Music - Virus synthesizer manufacturer
• Intelligent Dance Music - Genre background

Category:Machine Learning Category:Electronic Music Category:Audio Analysis Category:Synthesizers