Add ring-bench to compare SI-01 with other feedback suppressors #3

Open

erdoar wants to merge 4 commits from bench into main

erdoar commented

2026-10-15 05:10:58 +00:00

(Migrated from github.com)

Summary

Adds bench/, a separate Swift package (ring-bench) that tests feedback suppressors from the outside, using the same signals for each.
It generates 8 test signals: a latency sweep, three growing howls (one on a sung note's harmonic), and four feedback-free voice signals (vibrato, straight note, fricatives, fry).
It renders them through SI-01's engine offline, through any Audio Unit plug-in offline (for example Defeedback), or through live hardware.
It reports latency, time to a 6 dB cut, howl depth, and loss on everything other than the howl.
No plug-in internals are inspected; plug-ins are only hosted and run.

First results (main, offline, 256-frame blocks)

SI-01 leaves voices intact (95th-percentile loss of 1.5 dB or less) but barely touches howls. It cuts the 2.5 kHz howl by 3.7 dB and the 315 Hz howl by 13.7 dB, and it misses a howl on a sung note's harmonic entirely.

Verification

ring-bench selftest passes: the measures recover a known 37-sample delay with a -3 dB gain, and a known notch that engages 0.3 s after onset.
The AU path was checked with Apple's AUHipass.
Defeedback has not been rendered yet: it needs its installer run first.

🤖 Generated with Claude Code

## Summary - Adds `bench/`, a separate Swift package (`ring-bench`) that tests feedback suppressors from the outside, using the same signals for each. - It generates 8 test signals: a latency sweep, three growing howls (one on a sung note's harmonic), and four feedback-free voice signals (vibrato, straight note, fricatives, fry). - It renders them through SI-01's engine offline, through any Audio Unit plug-in offline (for example Defeedback), or through live hardware. - It reports latency, time to a 6 dB cut, howl depth, and loss on everything other than the howl. - No plug-in internals are inspected; plug-ins are only hosted and run. ## First results (main, offline, 256-frame blocks) SI-01 leaves voices intact (95th-percentile loss of 1.5 dB or less) but barely touches howls. It cuts the 2.5 kHz howl by 3.7 dB and the 315 Hz howl by 13.7 dB, and it misses a howl on a sung note's harmonic entirely. ## Verification - `ring-bench selftest` passes: the measures recover a known 37-sample delay with a -3 dB gain, and a known notch that engages 0.3 s after onset. - The AU path was checked with Apple's AUHipass. - Defeedback has not been rendered yet: it needs its installer run first. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

This pull request has changes conflicting with the target branch.

SI01/Support/Info.plist

View command line instructions

Manual merge helper

Use this merge commit message when completing the merge manually.

Merge commit title

Merge pull request 'Add ring-bench to compare SI-01 with other feedback suppressors' (#3) from bench into main

Merge commit body

Reviewed-on: https://dev.nulytica.com/forgeadmin/ring/pulls/3

Merge pull request 'Add ring-bench to compare SI-01 with other feedback suppressors' (#3) from bench into main

Reviewed-on: https://dev.nulytica.com/forgeadmin/ring/pulls/3

Checkout

From your project repository, check out a new branch and test the changes.

git fetch -u origin bench:bench

git switch bench

Merge

Merge the changes and update on Forgejo.

Warning: The "Autodetect manual merge" setting is not enabled for this repository, you will have to mark this pull request as manually merged afterwards.

git switch main

git merge --no-ff bench

git switch bench

git rebase main

git switch main

git merge --ff-only bench

git switch bench

git rebase main

git switch main

git merge --no-ff bench

git switch main

git merge --squash bench