feat(speaker): same-voice consolidation to cut over-segmented speakers#1140
Draft
r3dbars wants to merge 2 commits into
Draft
feat(speaker): same-voice consolidation to cut over-segmented speakers#1140r3dbars wants to merge 2 commits into
r3dbars wants to merge 2 commits into
Conversation
Adds a third EmbeddingClusterer post-process pass that agglomeratively merges large clusters whose mean embeddings clear the SpeakerNamingPolicy auto-accept bar (0.88), recomputing centroids after each merge so distinct speakers in crowded meetings don't chain-collapse. Fixes one remote voice surfacing as 4-7 "speakers" to name on a one-on-one call. - New consolidateSameVoiceClusters() wired into postProcess via optional consolidationThreshold (default 0.88; nil to skip) - Quality-filtered embeddings with all-sample fallback per cluster - Test coverage in EmbeddingClustererTests Resurrected from uncommitted WIP in the local checkout. Co-Authored-By: Claude Opus 4.8 (1M context) <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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Resurrected from uncommitted WIP that was sitting in the local main checkout (staged but never committed). Opened as a draft for review.
What it does
Adds a third
EmbeddingClustererpost-process pass — same-voice consolidation — that fixes over-segmentation: one real voice (often a remote participant under offline VBx clustering) split across several clusters that each accumulate enough speech to surviveabsorbSmallClusters. The symptom is a one-on-one call surfacing 4–7 "speakers" to name for a single person.consolidateSameVoiceClusters()agglomeratively merges clusters whose mean embeddings clear theSpeakerNamingPolicyauto-accept bar (0.88), recomputing centroids after each merge so the combined centroid must still clear the bar before the next cluster joins — avoiding the transitive A≈B, B≈C → A+B+C collapse that made broad pairwise merging unsafe on VBx output.postProcessvia an optionalconsolidationThreshold(default0.88, passnilto skip).qualityScore ≥ 0.3,duration ≥ 1.0) with an all-sample fallback so every cluster has a centroid.EmbeddingClustererTests.2 files, +233/-4. Branches cleanly off current
origin/main.#1114 ("Speaker-naming simulator + same-voice consolidation to cut speakers-to-name") tackles the same problem with a different implementation of
EmbeddingClusterer. These two are not identical and will conflict. Decide which approach wins (or reconcile) before merging either — don't land both blindly.🤖 Generated with Claude Code
Codex CI fix / review (2026-06-16)
886d6e28to fix the stale simulator negative-control fixture that failed Swift CI.SpeakerNamingSimulationRunnerTests/testSimulationReportFlagsConfusionFalseMergeAndFalseSplitlocally afterbash build-deps.sh --force; it now passes.bash build-deps.sh --force,bash build.sh --no-open,bash run-tests.sh,bash run-integration-smoke.sh,swift test, plus focusedSpeakerNamingSimulationRunnerTestsandEmbeddingClustererTests.~/.codex/skills/codex-review/scripts/codex-review --mode branchreported no accepted/actionable findings.DIRTYagainst currentorigin/mainwith conflicts inEmbeddingClusterer.swiftandEmbeddingClustererTests.swift. eval: multi-corpus speaker-naming harness + AMI scale-up #1144 is eval harness scale-up only, not an app-behavior duplicate.