Feature to unlink noise suppression for multi-channel inputs

[jacksongoode]

This alters a behavior that would have otherwise caused a quiet/noisy channel to apply noise suppression to a signal channel - current theory being this behavior preserves the stereo image. It adds a new feature experimental-unlink-ns that patches the problematic filter aggregation & cross-channel application.

If the intention in a multi-channel input is to capture, as an example, a stereo image from two mics, this feature might cause distortion in the representation as each channel will have its own noise filter applied.

However for non-linked microphones, dynamic gain levels can disrupt the performance of NS on a mic, therefore we make the offender (the Wiener filter aggregation) act on each channel individually.

The fix points to https://gitlab.freedesktop.org/jacksongoode/webrtc-audio-processing/-/commit/29ac3aab4df478b3f06698578c3a66f2ed2ac41e

[jacksongoode]

@strohel All ready for a look! And I think when this gets merged I'd be happy to cut 2.0.4 (unless you think this counts as a feature for a 2.1.0).

[strohel]

The approach looks very good, just a bunch of small technical details.

I also see that we haven't been documenting our feature flags properly in README.md - let's do it for this one (feel free to also document other missing flag there, up to you).

[jacksongoode]

@strohel I believe all your questions have been addressed, PTAL.

[strohel]

I've looked deeper and found some deeper problems (kinda), but should be quick to fix. 🙏

[jacksongoode]

Changes made to properly measure the RMS, patch with the dry reverse check, properly normalize i16 range, and better comments were added, asserts updated. @strohel have another look when time permits.

[strohel]

LGTM! I'll go ahead and cut a path release with this.

[strohel]

@jacksongoode ah one think I've realized post-merge:

if you enable the feature, build, disable the feature and build again, it stays enabled:

strohel@thicky ~/work/webrtc-audio-processing $ gs
On branch main
Your branch is up to date with 'origin/main'.

nothing to commit, working tree clean
strohel@thicky ~/work/webrtc-audio-processing $ cargo test --features experimental-unlink-ns
   Compiling webrtc-audio-processing-config v2.0.3 (/home/strohel/work/webrtc-audio-processing/webrtc-audio-processing-config)
   Compiling webrtc-audio-processing-sys v2.0.3 (/home/strohel/work/webrtc-audio-processing/webrtc-audio-processing-sys)
   Compiling webrtc-audio-processing v2.0.3 (/home/strohel/work/webrtc-audio-processing)
    Finished `test` profile [unoptimized + debuginfo] target(s) in 19.87s
     Running unittests src/lib.rs (target/debug/deps/webrtc_audio_processing-0f40afab18ddacbd)

running 12 tests
test tests::test_create_drop ... ok
test tests::test_process_signatures ... ok
test tests::test_zero_channels ... ok
test tests::test_full_aec_with_linear_aec_output_misconfiguration ... ok
test tests::test_nominal ... ok
test tests::test_tweak_processor_params ... ok
test tests::test_aec3_configuration_behavior ... ok
test tests::test_echo_cancellation_effectiveness ... ok
test tests::test_aec3_configuration_impact ... ok
test tests::test_unlink_ns_processing ... ok
test tests::test_stream_delay ... ok
test tests::test_nominal_threaded ... ok

test result: ok. 12 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 1.05s

     Running unittests src/lib.rs (target/debug/deps/webrtc_audio_processing_config-b5be71de2515ab82)

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

     Running unittests src/lib.rs (target/debug/deps/webrtc_audio_processing_sys-d444efa9de7d4db5)

running 5 tests
test tests::test_create_delete ... ok
test tests::test_empty_stats ... ok
test tests::test_config ... ok
test tests::test_process ... ok
test tests::test_some_stats ... ok

test result: ok. 5 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

   Doc-tests webrtc_audio_processing

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

   Doc-tests webrtc_audio_processing_config

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

   Doc-tests webrtc_audio_processing_sys

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

strohel@thicky ~/work/webrtc-audio-processing $ gs
On branch main
Your branch is up to date with 'origin/main'.

Changes not staged for commit:
  (use "git add <file>..." to update what will be committed)
  (use "git restore <file>..." to discard changes in working directory)
  (commit or discard the untracked or modified content in submodules)
        modified:   webrtc-audio-processing-sys/webrtc-audio-processing (modified content)

no changes added to commit (use "git add" and/or "git commit -a")
strohel@thicky ~/work/webrtc-audio-processing $ gd
strohel@thicky ~/work/webrtc-audio-processing $ cargo test --features bundled
   Compiling webrtc-audio-processing-sys v2.0.3 (/home/strohel/work/webrtc-audio-processing/webrtc-audio-processing-sys)
   Compiling webrtc-audio-processing v2.0.3 (/home/strohel/work/webrtc-audio-processing)
    Finished `test` profile [unoptimized + debuginfo] target(s) in 19.90s
     Running unittests src/lib.rs (target/debug/deps/webrtc_audio_processing-a2141e506ff1c077)

running 12 tests
test tests::test_create_drop ... ok
test tests::test_process_signatures ... ok
test tests::test_zero_channels ... ok
test tests::test_full_aec_with_linear_aec_output_misconfiguration ... ok
test tests::test_nominal ... ok
test tests::test_tweak_processor_params ... ok
test tests::test_aec3_configuration_behavior ... ok
test tests::test_echo_cancellation_effectiveness ... ok
test tests::test_aec3_configuration_impact ... ok
test tests::test_unlink_ns_processing ... FAILED
test tests::test_stream_delay ... ok
test tests::test_nominal_threaded ... ok

failures:

---- tests::test_unlink_ns_processing stdout ----

thread 'tests::test_unlink_ns_processing' (3783544) panicked at src/lib.rs:1132:13:
Expected ~-33.7dB without patch, got -22.866814
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace


failures:
    tests::test_unlink_ns_processing

test result: FAILED. 11 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 1.05s

error: test failed, to rerun pass `-p webrtc-audio-processing --lib`

Not a blocker for tagging, though it is suboptimal. At this point I think we should copy the source and only patch the copy.

[jacksongoode]

if you enable the feature, build, disable the feature and build again, it stays enabled:

Yeah, that's expected. It's again a case of the caching behavior. I suppose you mean copy the source to patch outside of the standard build cache location?

[strohel]

Right, though I wouldn't call it "caching". We modify the file in-place inside the git submodule.

Another problem I see is that we leave .rej files around if you build with the feature for the second time (cargo build --features experimental-unlink-ns; touch webrtc-audio-processing-sys/build.rs; cargo build --features experimental-unlink-ns).

meson has source directory and build directory. Currently we use /webrtc-audio-processing-sys/webrtc-audio-processing (the git submodule) as source directory directly. (and we patch it there, in-place).

The build directory is under OUT_DIR (target/debug/deps/webrtc_audio_processing_sys-d444efa9de7d4db5/).

I got nerd sniped and working on a PR.