Skip to content

AetherDSP Settings

AetherDSP Settings provides advanced control over AetherSDR's client-side noise-reduction engines: NR2, NR4, MNR, DFNR, RN2, and BNR. Each engine is selectable via a toggle row at the top of the dialog; clicking a toggle both selects that engine's page and activates or bypasses the engine.

Opening AetherDSP Settings

  1. Click Settings > AetherDSP Settings....
  2. The dialog opens.

Dialog controls

The AetherDSP Settings dialog uses a custom frameless chrome with a gradient title bar, minimize/maximize/close buttons, drag-to-move, and 8-axis resize. The dialog geometry is persisted and restored across sessions.

Control Behavior
Title bar โ€” AetherDSP Settings 18 px gradient title bar with grip glyph (โ‹ฎโ‹ฎ) on the left and the dialog title
โ€” (Minimize) Minimizes the dialog
โ–ก (Maximize) Maximizes or restores the dialog
ร— (Close) Closes the dialog
Drag-to-move Click and drag the title bar to move the dialog. Double-click to toggle maximize/restore
8-axis resize Click and drag any edge or corner to resize. Cursor changes to indicate direction. 6 px resize hit zone around the inner content widget

NR2 tab

The NR2 (musical-noise-reduction) engine uses a spectral-subtraction approach with configurable gain methods and noise power estimators.

Control Default Range Setting Key Behavior
Gain Method Gamma Linear, Log, Gamma, Trained NR2GainMethod Selects gain-curve mapping. Stored as integer 0-3
NPE Method OSMS OSMS, MMSE, NSTAT NR2NpeMethod Selects noise power estimator. Stored as integer 0-2
AE Filter (artifact elimination) Enabled On/Off NR2AeFilter Toggles the anti-artefact post-filter
Reduction: 1.50 0.50โ€“2.00 NR2GainMax Sets maximum NR2 reduction depth. Slider stores value*100 internally
Smoothing: 0.85 0.50โ€“0.98 NR2GainSmooth Controls how smoothly the noise estimate tracks changes
Threshold: 0.20 0.05โ€“0.50 NR2Qspp Sets speech-presence-probability threshold
Reset Defaults (โ†บ icon) โ€” โ€” โ€” Restores NR2 defaults: Gamma, OSMS, AE on, 1.50, 0.85, 0.20

NR4 tab

The NR4 engine uses the libspecbleach library for noise reduction. It offers configurable noise estimation methods and spectral processing controls.

Note: On Windows, NR4 requires LLVM (clang-cl) to be installed when compiling the source. If LLVM is not present, the NR4 toggle is dimmed and displays the tooltip "NR4 requires LLVM (clang-cl) on Windows. Install LLVM from llvm.org and rebuild to enable NR4."

Control Default Range Setting Key Behavior
Noise Estimation: MMSE MMSE, Brandt, Martin NR4NoiseEstimationMethod Selects noise-floor estimator. Stored as integer 0-2
Adaptive Noise Estimation Enabled On/Off NR4AdaptiveNoise Enables continuous re-estimation of the noise floor
Reduction (dB): 10.0 0.0โ€“40.0 NR4ReductionAmount Sets maximum NR4 noise reduction in dB. Slider stores value*10
Smoothing (%): 0 0โ€“100 NR4SmoothingFactor Time-domain smoothing of NR4 noise estimate
Whitening (%): 0 0โ€“100 NR4WhiteningFactor Flattens residual noise spectral shape
Masking Depth: 0.50 0.00โ€“1.00 NR4MaskingDepth Controls spectral-masking depth
Suppression: 0.50 0.00โ€“1.00 NR4SuppressionStrength Overall NR4 suppression strength
Reset Defaults (โ†บ icon) โ€” โ€” โ€” Restores NR4 defaults: MMSE, adaptive on, 10 dB, 0, 0, 0.50, 0.50

MNR tab (macOS only)

The MNR (macOS MMSE-Wiener) engine is available only on macOS builds. It provides asymmetric gain smoothing for noise reduction. The MNR toggle is dimmed on Windows and Linux builds โ€” the engine has no backend on those platforms.

Control Default Range Setting Key Behavior
Enable MNR (macOS only) Disabled On/Off MnrEnabled Enables MMSE-Wiener noise reduction with asymmetric gain smoothing. Initial state read live from AudioEngine
Strength 100 0โ€“100 MnrStrength Adjusts MNR aggressiveness (0 mild, 100 max). Persisted as normalized 0.00โ€“1.00

DFNR tab

The DFNR (DeepFilterNet3) engine uses a deep learning model for noise reduction.

Control Default Range Setting Key Behavior
Attenuation Limit 100 0โ€“100 dB DfnrAttenLimit Sets maximum noise attenuation (0 = passthrough, 100 = maximum)
Post-Filter Beta 0.00 0.00โ€“0.30 DfnrPostFilterBeta Applies an additional post-filter for extra suppression. Slider stores value*100 internally

RN2 tab

The RN2 (RNNoise) tab is purely informational and has no adjustable parameters.

BNR tab

The BNR (NVIDIA) tab's intensity is controlled from the overlay menu. The BNR toggle is dimmed on builds without the NVIDIA Broadcast SDK.

Engine selection and mutual exclusion

The six DSP toggles (NR2, NR4, MNR, DFNR, RN2, BNR) act as both page selectors and engine enable/disable controls. When NR2 is activated, the AudioEngine cascades exclusion, disabling DFNR and other mutually exclusive modules. Only one engine can be active at a time.

Tips

  • Masking Depth: and Suppression: on the NR4 tab interact: raising both together produces maximum noise reduction but the highest risk of speech distortion. Raise them incrementally and test on a live or recorded signal.
  • If speech sounds over-processed or hollow, reduce Masking Depth: first, then Suppression: until naturalness returns.
  • The Adaptive Noise Estimation checkbox affects how quickly NR4 tracks a changing noise floor, which in turn affects how both sliders sound in practice.
  • Click Reset Defaults on any tab to return all parameters on that tab to their factory values.

Troubleshooting

  • Speech sounds hollow or underwater after raising the sliders โ€” Both sliders at high values can over-suppress spectral components that overlap with speech. Reduce Masking Depth: first, then Suppression: until naturalness returns.
  • Noise floor is still audible even at maximum settings โ€” Ensure Adaptive Noise Estimation is enabled so NR4 can continuously re-estimate the noise floor. Also consider increasing Reduction (dB): .
  • Slider snaps back or refuses to move โ€” Click directly on the slider handle rather than clicking in the groove.
  • NR4 toggle is dimmed on Windows โ€” The NR4 engine requires LLVM (clang-cl) to compile its C99 VLAs. Install LLVM from llvm.org and rebuild AetherSDR to enable NR4.
โšก Athena AetherSDR Assistant
Hi! I'm Athena, the AetherSDR AI assistant. Ask me anything about installation, configuration, or troubleshooting.