data(delivery-003): regen post-#107/#108/#109 — single_backend false positive resolved

[shaypal5]

Summary

Schema-shift regen of delivery-003 using SynthBanshee 1ea48f3 — tip of main after PRs #110, #111, #112 merged earlier today. All 20 clips regenerated; qa-report.json no longer fires the misleading single_backend warning that was filed as the open QA finding in the original delivery.

Why a regen (option 1) rather than in-place post-processing (option 2)

Two motivations made the regen worthwhile:

1. Audio cost was effectively $0. Azure clips (18 of 20) hit the SHA-256 SSML cache and re-rendered byte-identical WAVs. The two Google Chirp HD clips re-rendered, costing fractions of a cent. Total wall time: ~16s for both batches.
2. Highest fidelity guarantee. The regen produces canonical artifacts straight from the pipeline; no manual JSON editing or path-string sed-ing. The generator is now the source of truth for what delivery-003 should look like under the post-#109 schema.

Schema changes vs the original 2026-05-12 commit

Field	Before	After
tts_engine (clip JSON)	always "azure_he_IL" (wrong for the 2 Google clips)	field absent — Pydantic drops it; backend is generation_metadata.tts_backend per speaker (#112)
dirty_file_path (clip JSON)	absolute pytest tmp_path on sp_neu_a_0001_00 (#107 fingerprint); empty/absent on others	repo-relative POSIX assets/speech/dirty/... everywhere (#110, #111)
transcript_path (clip JSON)	corpus PR #4 had post-processed to relative	now produced relative by the generator itself (#111)
manifest.csv wav_path / strong_labels_path	corpus PR #4 had post-processed	now produced relative (#111)
qa-report.json run_summary.clips_by_tts_engine	{"azure_he_IL": 20}	renamed clips_by_tts_backend with {"azure": 18, "google": 2} (#112)
qa-report.json run_summary.run_warnings	included single_backend (false positive)	single_backend resolved; only legitimate low_voice_diversity_* warnings remain

Audio integrity

• 19 of 20 clips: WAV bytes byte-identical with the original delivery-003 (Azure SSML cache hit). Only metadata JSON changed.
• sp_sv_a_0003_00.wav (Google Chirp HD): bit-level difference from the original render — Google doesn't share the content-hash cache the same way Azure does in this codebase. Validation still PASSED (16 kHz, mono, peak ≤ −1.0 dBFS, duration ≥ 3 s). Total corpus duration shifted by 2.2s (2500.79 → 2498.63 s).

Files changed per area

Area	Count	Notes
Per-clip JSON (data/he/**/*.json)	20	Schema migration
Google clip audio + transcript + JSONL (sp_sv_a_0003_00.{wav,txt,jsonl})	3	Re-render
data/he/manifest.csv	1	Relative paths; 20 rows; all split: train
assets/speech/*.wav (new SSML caches)	6	From the Google re-render
deliveries/003-multi-project-multi-voice/qa-report.json	1	New schema; single_backend gone
deliveries/003-multi-project-multi-voice/metadata.yaml	1	Pinned new commit; added regen_2026_05_12 + qa_findings_closed_post_regen_2026_05_12 sections
deliveries/003-multi-project-multi-voice/notes.md	1	Pipeline section split into "initial" + "schema-shift regen"; closed-findings list refreshed
DELIVERIES.md	1	Added #110/#111/#112 to pipeline-milestone column

Test plan

• synthbanshee qa-report data/he --output deliveries/003-multi-project-multi-voice/qa-report.json --run-summary — failure rate 0.0%, 20/20 clips passed, no single_backend warning
• synthbanshee validate spot-checked on sp_sv_a_0003_00.wav (Google re-render) and el_sv_b_0001_00.wav (Azure cache hit) — both VALID
• jq spot-checks on 3 sampled JSONs confirm has_tts_engine: false, repo-relative transcript_path and dirty_file_path, populated generation_metadata.tts_backend
• Manifest CSV wav_path column verified repo-relative for all 20 rows
• Audio total duration delta vs original (2498.63 vs 2500.79) — accounted for by sp_sv_a_0003_00 Google re-render

Tier-3 ASR sanity (local)

Not applicable. The three SynthBanshee PRs that prompted this regen touch tests/, synthbanshee/cli.py path-shape, synthbanshee/package/manifest.py, synthbanshee/package/qa.py, synthbanshee/labels/, and docs — none touch synthbanshee/tts/, synthbanshee/script/, synthbanshee/augment/, nor any speaker / scene / acoustic / project YAML config. The audio pipeline itself is unchanged, and the Azure cache hits prove it (19/20 bit-identical WAVs). Per SynthBanshee CLAUDE.md "ASR sanity check policy", the regen is exempt from the local `qa-report --asr` run.

🤖 Generated with Claude Code

[github-actions]

pr-agent-context report:

No unresolved review comments, failing checks, or actionable patch coverage gaps were found on PR #5 in repository https://github.com/DataHackIL/avdp-synth-corpus. Treat this PR as all clear unless new signals appear.

Run metadata:

Tool ref: v4
Tool version: 4.0.21
Trigger: commit pushed
Workflow run: 25758545236 attempt 1
Comment timestamp: 2026-05-12T19:54:16.196657+00:00
PR head commit: 3e6335ea75109a41594fb9fcf590f109b27d10c6