Version 0.4.2
FIXED-RULE AUDIO WORKFLOW FOR MIXING AND MASTERING MUSIC STEMS
Designed for amateur music producers and hobbyists
Overview • Feature Set • First Session • Quick Start • System Requirements • Build + Install • Licensing
AutoMixMaster helps you turn raw stems into a cleaner, release-ready track with fewer manual steps.
It focuses on predictable, fixed-rule processing so results are repeatable and beginner-friendly, with optional extras like AI stem separation, batch mode, and bundled renderer integrations.
Core capabilities at a glance
| Module | Description |
|---|---|
| Auto Mix + Auto Master | Deterministic stem balancing, gain staging, and limiting workflow. |
| One-Click Pipeline | Mix + Master (Ctrl+Shift+M) runs Auto Mix → Auto Master → Export. |
| ITO-Master AI Mastering | Experimental AI mastering strategy driving a 46-parameter native white-box FX chain with licensing consent gating. |
| AI Stem Separation (Optional) | Splits a single full-mix import into stems before processing when enabled (Ctrl+Shift+A). |
| Task-Scoped Model Browser | Install/uninstall models and set active packs per task (mix, master, analysis, separation) from Hugging Face or GitHub Releases. |
| Batch Processing | Queue folders, auto-group stems by filename role patterns, and render one mastered song per group (<song>_AutoMixMaster_YYYYMMDD_XX.<ext>). Supports recursive discovery via UI toggle or AUTOMIX_BATCH_RECURSIVE=1. |
| Renderer Integrations | Built-in discovery for PhaseLimiter, FFmpeg, SoX, and rsgain; only available tools are shown (*_BIN env overrides supported). |
| Verification Reporting | Export verification report plus batch completion summary; optional per-export .report.json sidecar. |
| Task Center + ETA | Real-time progress tracking with batch ETA countdown, summary status row, timestamped activity log, and copy-log utility. |
| Transport & Audio Preview | Realtime-safe lock-free audio preview buffer, live peak/RMS meters, 0–1.5x gain control (+3.5 dB), and modal confirmation for session clearing. |
| Shortcuts & Commands | ApplicationCommandManager integration with built-in Keyboard Shortcuts Cheatsheet modal (?). |
| Analysis Meters | Live LUFS and peak metering via GlowMeters. |
Welcome to your first mixing and mastering session. AutoMixMaster simplifies the process into a few core steps:
- Import audio: Drag and drop files onto the waveform area, or click
Import. Supported formats: WAV, AIFF, FLAC, MP3, OGG. - (Optional) enable AI Stem Separation: Toggle AI Stem Separation and check the badge beside it (
Separation model: <name/none>).- If exactly one full-mix track is loaded, separation runs before Auto Mix.
- If multiple files are loaded, they are treated as regular stems and separation is skipped.
- Manage models in Model Browser: Open Models to fetch catalog entries, install/uninstall models, and set active packs per task (
mix,master,analysis,separation) using Set Active or Use Selected for Task. - Auto Mix: Click Auto Mix to analyze stems and apply deterministic balancing rules.
- Auto Master: Click Auto Master to apply mastering strategy and limiting.
- One-click pipeline: Click Mix + Master (
Ctrl+Shift+M) to run Auto Mix → Auto Master → Export. If AI Stem Separation is enabled and one full mix is loaded, separation is performed first, then the pipeline continues automatically. - Export: Use Export (
Ctrl+E) for manual output control, or rely on pipeline export.
Getting started with AutoMixMaster is simple.
⚠️ Testing disclaimer: only the Windows version has been manually tested end-to-end so far.
Linux, macOS, and ARM64 artifacts are currently provided as best-effort builds.
Windows users can download the portable release zip (AutoMixMaster-windows-<arch>.zip), extract it, and run AutoMixMaster.exe.
macOS users can download the release zip (AutoMixMaster-macos-<arch>.zip), extract it, and open AutoMixMaster.app.
Linux users can download either:
- AppImage (
AutoMixMaster-<version>-<arch>.AppImage) for a portable one-file launch. - Debian package (
automixmaster_<version>_<arch>.deb) for Ubuntu/Debian install. - Flatpak bundle (
AutoMixMaster-linux-<arch>.flatpak) for Flatpak-based installs.
If you are on Linux, or prefer to build the application from source on Windows, refer to the Build + Install section below for verified instructions.
These are practical estimates for AI-heavy workflows (especially ONNX-based separation/mix/master inference), not strict hard limits.
AutoMixMaster is designed to benefit from GPU acceleration via ONNX Runtime providers.
- Windows: Windows 10 or Windows 11 (x64 or ARM64)
- macOS (Apple Silicon / ARM64): macOS 14+
- macOS (Intel / x64): macOS 15+
- Linux: Ubuntu 24.04 LTS+ for current prebuilt
.deb/AppImage artifacts
Note: Ubuntu 22.04 may still work if you build from source on 22.04 with compatible dependencies, but official CI/release packaging currently targets Ubuntu 24.04.
- CPU: modern 6-core / 12-thread desktop CPU (Ryzen 5 5600 / Core i5-12400 class)
- RAM: 16 GB minimum
- GPU: compatible acceleration path with ~6 GB VRAM
- Windows DirectML path: DirectX 12-capable GPU
- CUDA path: NVIDIA CUDA-capable GPU
- Storage: ~10 GB free (models, temp files, exports)
- CPU: 8 cores / 16 threads or better (Ryzen 7 / Core i7 class)
- RAM: 32 GB
- GPU: 8–12 GB VRAM
- CPU: 12 cores+ strongly recommended
- RAM: 32–64 GB
- GPU: 12 GB+ VRAM
- GPU acceleration matters most: DirectML needs a DirectX 12 GPU and CUDA needs an NVIDIA CUDA-capable GPU (DirectML, CUDA).
- Demucs notes roughly 3 GB minimum and around 7 GB typical GPU memory, so 8 GB+ VRAM is a safer real-world target; CPU-only runs work but are slower (Demucs README).
- Configure
cmake -S . -B build -G "Visual Studio 18 2026" -A x64- Build
cmake --build build --config Release --parallel- Install dependencies
sudo apt-get install -y \
libasound2-dev libfreetype6-dev libx11-dev libxcomposite-dev \
libxcursor-dev libxext-dev libxinerama-dev libxrandr-dev \
libxrender-dev libwebkit2gtk-4.1-dev libglu1-mesa-dev mesa-common-dev- Configure + build
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build --parallel- Run tests (optional but recommended)
ctest --test-dir build --output-on-failure- Install build tools
xcode-select --install
brew install cmake ninja- Configure + build (pick one architecture)
# Apple Silicon (arm64)
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DCMAKE_OSX_ARCHITECTURES=arm64 -DBUILD_TESTING=OFF -DBUILD_TOOLS=OFF
# Intel (x64)
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DCMAKE_OSX_ARCHITECTURES=x86_64 -DBUILD_TESTING=OFF -DBUILD_TOOLS=OFF
cmake --build build --target AutoMixMasterApp --parallel- Configure + build (universal binary: arm64 + x86_64)
cmake -S . -B build-universal \
-DCMAKE_BUILD_TYPE=Release \
-DCMAKE_OSX_ARCHITECTURES="arm64;x86_64" \
-DBUILD_TESTING=OFF \
-DBUILD_TOOLS=OFF
cmake --build build-universal --target AutoMixMasterApp --parallel- Install app bundle
APP_BUNDLE="$(find build build-universal -maxdepth 6 -type d -name 'AutoMixMaster.app' 2>/dev/null | head -n 1)"
cp -R assets "$APP_BUNDLE/Contents/MacOS/assets"
sudo cp -R "$APP_BUNDLE" /Applications/
open /Applications/AutoMixMaster.appAfter building, create distributable Linux packages with:
./tools/package_linux.shOutput artifacts are written to dist/linux/.
Manifest path:
packaging/flatpak/io.automixmaster.AutoMixMaster.yml
Note: the Flatpak manifest prefetches JUCE/nlohmann/libebur128 sources and passes
FETCHCONTENT_SOURCE_DIR_*flags so CMake does not need live GitHub access inside the Flatpak sandbox.
Install Flatpak tooling:
sudo apt-get install -y flatpak flatpak-builderAdd Flathub and install required runtime/SDK:
flatpak remote-add --user --if-not-exists flathub https://flathub.org/repo/flathub.flatpakrepo
flatpak --user install -y flathub org.freedesktop.Platform//24.08 org.freedesktop.Sdk//24.08Build bundle:
./tools/build_flatpak.shOutput:
dist/flatpak/AutoMixMaster.flatpak
The renderer registry can auto-discover optional CLI tools if you place binaries under:
assets/ffmpeg/bin/ffmpeg(.exe)or setFFMPEG_BINassets/sox/bin/sox(.exe)or setSOX_BINassets/rsgain/bin/rsgain(.exe)or setRSGAIN_BIN
If a tool is missing, it is hidden from selectable available renderers automatically.
AutoMixMaster is distributed under the GNU General Public License v3 (GPLv3).
| Component | License | Role |
|---|---|---|
| JUCE 8.0.8 | AGPLv3 / Commercial | Audio & GUI Framework |
| libebur128 | MIT | EBU R128 Loudness Metering |
| nlohmann/json | MIT | JSON Serialization & Model Metadata |
| Catch2 3.7.1 | BSL-1.0 | Unit Testing Framework |
| PhaseLimiter | GPL-2.0 / Custom | Optional External Limiter |
| FFmpeg | GPL-compatible / LGPL | Optional External Audio Renderer |
| SoX | GPL-2.0-or-later | Optional External Processor |
| rsgain | BSD-2-Clause | Optional ReplayGain Tagging Stage |
Model weights are not bundled into the installer or executable binaries. Users can optionally download models on-demand through the built-in Model Hub (ModelManager), which preserves and displays upstream model licensing metadata (license / cardData tags):
| Model / Model Family | Upstream Author | License | Usage & Compatibility |
|---|---|---|---|
| HTDemucs / Demucs 4-stem | Meta Research | MIT (code), CC-BY-NC 4.0 (MUSDB18-HQ weights) | Stem Separation (Non-Commercial weights) |
HTDemucs 6-stem (htdemucs_6s) |
Meta / Community ONNX | CC-BY-NC 4.0 | 6-Stem Separation (Guitar/Piano/Drums/Bass/Vocal/Other) |
Denoiser (dns64 / Speech Enhancer) |
Meta Research | CC-BY-NC 4.0 | Speech & Vocal Denoising (Non-Commercial evaluation) |
ITO-Master (ito-master-v1) |
Community / Open | CC-BY-NC 4.0 | AI Mastering (46-param native white-box FX chain) |
| Whisper Tiny / Small | OpenAI | MIT | Transcripts, Vocal Alignment & Pitch Analysis |
CLAP (clap-htsat) |
LAION | MIT | Style Retrieval & Audio Embeddings |
PANNs (PANNs_CNN14) |
Bio-DSP / Open | MIT | General Audio Tagging & Classification |
Non-Commercial Notice: Models licensed under CC-BY-NC 4.0 (such as Meta Demucs, Denoiser, and ITO-Master weights) are restricted to personal, educational, and non-commercial evaluation use. Commercial workflows can use open-source MIT-licensed models (e.g. Whisper, CLAP) or the built-in deterministic heuristic DSP engines. User consent gating is enforced prior to model download and execution.
Model Licensing Audit: For complete machine-checkable model license metadata and audit reports, see docs/model-licensing-audit.md and docs/model-licensing-audit.json.