Tune NR2 Reduction Depth and Voice Threshold¶
This page explains how to adjust NR2's maximum noise reduction depth and speech-presence threshold in AetherSDR. Changing these two sliders lets you balance how aggressively NR2 suppresses noise against how faithfully it preserves speech.
Before you start¶
- AetherSDR must be running. A radio connection is not required to change these settings.
- NR2 must already be active on a receiver slice for changes to take audible effect immediately.
Steps¶
- Click
Settings > AetherDSP Settings.... - In the AetherDSP Settings dialog, click the NR2 tab.
- Locate the Reduction Depth: slider. Drag it left to reduce suppression or right to increase it. The current value is shown to the right of the slider (default:
1.50). - Locate the Voice Threshold: slider. Drag it left to make speech detection more sensitive (preserves quiet speech) or right to raise the threshold (passes less noise during pauses). The current value is shown to the right of the slider (default:
0.20). - Changes take effect immediately and are saved automatically.
To restore both sliders to their defaults along with all other NR2 parameters, click Reset Defaults at the bottom right of the NR2 tab.
What each control does¶
| Control | Default | Valid Range | Persisted Key | Behavior |
|---|---|---|---|---|
| Reduction Depth: | 1.50 |
0.50β2.00 |
NR2GainMax |
Sets the maximum depth of NR2 noise reduction. Higher values suppress more noise; values above 1.50 risk distorting speech. |
| Voice Threshold: | 0.20 |
0.05β0.50 |
NR2Qspp |
Sets the speech-presence-probability threshold. Lower values treat more signal as speech and preserve it; higher values treat more signal as noise and suppress it. |
| Smoothing: | 0.85 |
0.50β0.98 |
NR2GainSmooth |
Controls how quickly the noise estimate tracks changes. Higher values give a steadier estimate but slower adaptation to changing noise. |
| AE Filter (artifact elimination) | On | On / Off | NR2AeFilter |
Toggles the anti-artifact post-filter that reduces musical noise and ringing. |
| Gain Method | Gamma | Linear / Log / Gamma / Trained | NR2GainMethod |
Selects the gain-curve mapping used when computing suppression. |
| NPE Method | OSMS | OSMS / MMSE / NSTAT | NR2NpeMethod |
Selects the noise power estimator. |
| Reset Defaults | β | β | β | Restores all NR2 tab parameters to defaults: Gamma, OSMS, AE on, 1.50, 0.85, 0.20. |
Tips¶
- For SSB voice operation, start with Reduction Depth: at
1.50and Voice Threshold: at0.20. If speech sounds clipped or hollow, lower Reduction Depth: toward1.00. - Lowering Voice Threshold: below
0.15can cause residual noise to break through during speech pauses because more of the signal is classified as speech. Raise it if you notice this. - If the noise estimate reacts too slowly to burst noise, lower Smoothing: toward
0.60. If the noise gate sounds choppy, raise it toward0.95. - Leaving AE Filter (artifact elimination) enabled is recommended for most conditions; disable it only if you notice the post-filter itself introducing artifacts.
Troubleshooting¶
- Speech sounds hollow or over-processed β Reduction Depth: is too high or Voice Threshold: is too high. Lower Reduction Depth: and/or lower Voice Threshold: so more speech components are preserved.
- Noise is still audible during speech pauses β Voice Threshold: is too low, causing pauses to be classified as speech. Raise Voice Threshold: toward
0.30β0.40. - Noise estimate reacts sluggishly or the noise floor sounds unstable β Adjust Smoothing: (see Tips above). Also verify the selected NPE Method suits your noise type; NSTAT adapts better to non-stationary noise.