Skip to content

P0 Harmonice M3: preserve PSOLA overlap across octave readout errors and real vocal leaps - #548

Merged
masarray merged 3 commits into
integration/harmonice-lead-ensemble-20261111from
fix/harmonice-period-continuity-20261111
Nov 10, 2026
Merged

masarray merged 3 commits into
integration/harmonice-lead-ensemble-20261111from
fix/harmonice-period-continuity-20261111

Conversation

@masarray

Copy link
Copy Markdown
Owner

User field evidence

Latest Windows FAT for #529 significantly reduced but did not fully eliminate short harmony crackle and stutter. LOW is audible; busy chorus and sustained vibrato are still acceptance-critical, especially HIGH.

Root-path audit / precise change

The last HarmonyDsp::finishPitchAnalysis path still hard-reset all 5 pitchSync_ voice grains whenever the incoming source period differed by >25% from the previously tracked F0. Two analysis windows can confidently report an octave harmonic during a loud vowel; such a reset replaces all active singer grains by a fresh renderer and leaves dropout holes while it reacquires stable epochs. This is in the audio thread, even though PCM processing itself is realtime-safe.

  • Introduce one small fixed-state HarmonyPeriodContinuity.h policy: 1–2 large F0 period outliers are ignored, 3 consistent analyses admit a real new period, and source-period motion is capped at 11% per analysis hop without pitchSync_.reset(). Small genuine vibrato continues using the existing bounded 5% / 0.30 rate logic.
  • Guard large backing melodic leaps (>=9 semitones) with 4 coherent analysis windows, retaining the existing 2-window admission for normal sung note transitions and phrase onset.
  • Preserve existing native graph, JUCE callback ownership, 1536-sample @48k PDC, dry lead/correction default, stereo roles, GUI and current presets.
  • Add native C++ test for two-frame false octave errors, bounded monotonic true octave convergence, invalid evidence rejection, reset on new phrase. Existing detailed 5-voice loud-vibrato and per-block PCM parity regressions remain mandatory.

Admission plan

This is a focused DSP-only draft stacked on #529 and not on production main. Run portable CI first; then merge only to #529 candidate after evidence. Use the new scoped Windows FAT once per qualified source SHA; no Windows churn for speculative edits. Real singer A/B and physical ASIO XRUN remain open. Do not interpret synthetic pass as zero crackle in a live vocalist.

@masarray
masarray marked this pull request as ready for review November 10, 2026 23:05
@masarray
masarray merged commit 7496fee into integration/harmonice-lead-ensemble-20261111 Nov 10, 2026
3 of 4 checks passed
@chatgpt-codex-connector

Copy link
Copy Markdown
Codex Review Summary: ❌ Failed

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ❌ Failed 2026-11-10T23:05:08.608555Z
We couldn’t complete this review. Comment @codex review to try again.
7dbc311 Draft marked ready
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@chatgpt-codex-connector

Copy link
Copy Markdown

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant