Stage 16 audition prep: render example and purity characterization - #43
Open
robmorgan wants to merge 1 commit into
Open
Stage 16 audition prep: render example and purity characterization#43robmorgan wants to merge 1 commit into
robmorgan wants to merge 1 commit into
Conversation
…tion examples/stage16_audition.rs renders the blind-listening set for both Stage 16 questions: 3 corpus excerpts x +/-4/8% x 4 arms — the shipped SOLA path, PV-512/PV-1024 prototypes behind the 120 Hz split (the corrected-range re-audition on the phase-fixed vocoder), and a Rubber Band reference — RMS-matched into target/stage16_audition/. tests/tonal_purity_characterization.rs pins the measured granulation floor on music-like probes (characterization floors, not quality gates), with a finding the sine probe missed: the floor is strongly asymmetric. Harmonic-15 purity measures 52 dB at +8% tempo but 22 dB at -8% — slowdowns re-cross material so splices recur through sustained content — while two-tone beating survives well both ways (47/42 dB). The Stage 16 listening should focus on slowdowns. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Makes the Stage 16 listening session pure listening — everything up to the verdict is prepared.
examples/stage16_audition.rsRenders the blind set for both Stage 16 questions: 3 corpus excerpts (hot_stuff strings, msbwy filtered strings, cold_heart vocal) × ±4/±8% × 4 arms, RMS-matched, into
target/stage16_audition/:ours_sola.wav— the shipped keylock pathpv512.wav/pv1024.wav— the corrected-range re-audition prototypes: the keylock's exact two-band topology (LR8 @ 120 Hz, low band pitch-follows) with a fixed identity-locked PV as the high-band corrector — the Stage 7 experiment re-run on the phase-fixed vocoderrubberband.wav— R3 reference (CLI detected automatically)tests/tonal_purity_characterization.rsPins the measured granulation floor on music-like probes (characterization floors, not quality gates), with a finding the review's sine probe missed: the floor is strongly asymmetric.
Slowdowns re-cross material — the elastic cursor splices recur through sustained content — so the listening should focus on −4/−8%.
Listening protocol (Stage 16 exit)
Per track/rate folder: blind-compare
ours_solavsrubberband(granulation verdict: accept/document vs pursue) andpv512/pv1024vsours_solaon sustained content (does the July "phasey at every boundary" verdict survive the Stage 13 fixes?). Verdicts recorded in the ROADMAP stage note per the evidence-caveat protocol, implementation state alongside.🤖 Generated with Claude Code