Feature/ainr dev - #452
Open
vis4573 wants to merge 3 commits into
Open
Conversation
The spectral-subtraction DSP stage computed scaled_noise_threshold from the THSHLD field but never used it: the reduction sigmoid used a hardcoded midpoint of 0.5, so adjusting THSHLD had no audible effect at any setting. - src/sbitx.c: use scaled_noise_threshold as the sigmoid midpoint in the spectral-subtraction reduction factor. - src/sbitx_gtk.c: rescale scaleNoiseThreshold() output from 0.001-0.01 to 0.1-2.0. The old range was ~50x too small to shift the midpoint meaningfully against typical per-bin SNR values (SNR ~ 1.0 at the noise floor). New mapping: 0 = aggressive subtraction, 100 = conservative. Verified numerically (reduction curve now spreads properly across the slider range) and on-air on 40m SSB (THSHLD audibly changes subtraction aggressiveness).
Adds a third selectable noise-reduction mode alongside DSP and ANR: AINR, based on Xiph.org's RNNoise - a small GRU recurrent neural network trained for real-time speech denoising. Unlike the statistical DSP/ANR stages it classifies audio by learned speech structure (harmonics, pitch continuity, envelope), letting it suppress noise between words to near silence and handle non-stationary noise (static crashes, clicks) that power-tracking methods miss. User-facing: - AINR toggle on the main screen (left of REC); console \ainr on|off. - While ON, two tuning fields appear beside it: AINRS (0-100, default 80) - strength as a wet/dry mix; AINRV (0-50, default 25) - how far suppression relaxes while speech is detected, protecting weak/fading syllables from being gated. - Voice modes only; deliberately not applied in FT8/FT4/DIGITAL. Modem decoders always receive unprocessed audio (hook is after modem_rx). - Note: remote/web audio receives AINR-processed audio (the stage runs before the remote audio queue), so remote listeners benefit too. Implementation: - rnnoise/: vendored xiph/rnnoise v0.1.1 (BSD-3-clause, COPYING included). This version embeds the trained model weights in source (no network fetch at build time) and its FFT symbols are prefixed rnn_fft_*, so no collision with ft8_lib's kiss_fft. - src/sbitx_rnn.c/h: 96k<->48k conversion, ring-buffered reframing into 480-sample frames, automatic input level normalization toward the model's trained operating range (results independent of IF gain setting), VAD-adaptive wet/dry mixing with fast-attack/slow-release smoothing, sample-aligned dry path. ~10-20 ms added latency. - src/sbitx.c: rnn_enabled hook in the voice-mode post-processing path (after modem_rx, before RX EQ); state reset on toggle-off. - src/sbitx_gtk.c: field definitions, periodic value pickup in check_plugin_controls, conditional AINRS/AINRV visibility. - Makefile: rnnoise sources and include path; ./build sbitx unchanged. Testing: CPU on a Pi 4 sBitx v3 stays under 20% across all four cores with AINR active. Amplitude passthrough verified at exactly 1.0 for tone content; ~23x noise suppression at full strength; suppression quality verified independent of input level; VAD relax/recovery timing verified (~200 ms). Parameters tuned from multiple on-air sessions on 40m SSB.
Plain-language and technical writeups as a new README section: what AINR is, where the controls are, adjustable parameters (AINRS/AINRV) with ranges, defaults, and tuning guidance.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
I've been experimenting with adding neural-network noise reduction (AINR) to the sBitx firmware — it uses Xiph.org's RNNoise, a small AI model trained on thousands of hours of speech, so instead of just filtering frequencies it actually recognizes what a human voice sounds like and mutes everything else.
Results on 40m SSB have been impressive — band hiss between words drops to near silence while the voice stays clean. Works alongside the existing DSP/ANR options (use one at a time).
Two on-screen controls: AINR (on/off, main screen next to REC) and strength/weak-signal tuning. Runs fine on the stock Pi, ~15ms latency, voice modes only (won't touch your FT8).
Code is on my fork of the drexjj 64-bit firmware if anyone wants to try:
github.com/vis4573/sbitx (branch: feature/ainr-dev)
Build with the usual ./build sbitx. Feedback welcome — especially reports on weak-signal behavior!
73, VU3UBP
vu3ubp@gmail.com