[LauraGPT]

Final Publication Acceptance

GitHub Release and PyPI 1.4.26 are published. Both official PyPI distributions were downloaded and match GitHub by SHA-256 and size:

• funasr-1.4.26-py3-none-any.whl: 1015738 bytes; SHA-256 104e3704fed0ecfd6c08836468de147a2c6f05745281093b8a6dbe341d1ce831.
• funasr-1.4.26.tar.gz: 1086172 bytes; SHA-256 763c7d860a49afd17e4b9c0636ff43f03f6519e68bd0a09a25d68318091e2cf8.

FunASR 1.4.26

Source: cdd82b5cbfd05015f8fb2b167d61f25142c3a9e8 / tag v1.4.26.

Fixes

• fix(server): return plain text for response_format=text #3874: response_format=text now returns the bare HTTP transcript instead of a JSON-quoted string.
• fix(postprocess): keep a space between Latin VAD segments in rich transcription #3872 / fix(postprocess): preserve VAD word boundaries with current-main CI #3883: retain word separators across English and Korean VAD segments while preserving CJK and event-only behavior. The original author's commit is retained.

The independent ONNX/Torch abbreviation and batch-padding timestamp fixes are not shipped by this root wheel. Their package releases are separate. Open #3861, #3862, #3867 and the new hotword PRs are not claimed resolved.

Verification Scope

• fix(server): return plain text for response_format=text #3874 previously passed 32 actual HTTP/SDK cases and 213 related CPU cases; its source bytes remain unchanged here.
• fix(postprocess): preserve VAD word boundaries with current-main CI #3883 passed 246 unique selected CPU cases across the initial 230 passes and completion of six optional-dependency cases plus ten actual installed standalone-wheel cases. All original skipped and negative runs remain preserved. NumPy hosted CI has 421 passes and 27 CUDA-only skips per version, not GPU acceptance.
• Release/docs candidate checks: {'contract-green': {'testcase': 53, 'failure': 0, 'error': 0, 'skipped': 0}, 'static-remainder': {'testcase': 104, 'failure': 0, 'error': 0, 'skipped': 2}, 'installed-behavior-candidate': {'testcase': 111, 'failure': 0, 'error': 0, 'skipped': 0}, 'site-boundary-integration': {'testcase': 69, 'failure': 0, 'error': 0, 'skipped': 0}, 'site-manifest-integration': {'testcase': 21, 'failure': 0, 'error': 0, 'skipped': 0}}. Actual installed root wheel final counts: {'testcase': 111, 'failure': 0, 'error': 0, 'skipped': 0}. Standalone runtime paths are explicitly deselected from root-wheel tests. All 439 package files and 13 entrypoints match accepted source.
• Root model/GPU-runtime bytes are unchanged from 1.4.25 and remain bound to its accepted prior L20Z evidence. No new GPU execution, pretrained ASR-quality, full-suite or Nano result is claimed.
• Head, merged-main and final-tag hosted CI and actual checkout trees are separately accepted. Final Python wheel and sdist are newly downloaded and checked for metadata, exact source bytes, archive integrity, strict Twine, private offline installation and CLI behavior.
• Ten unchanged runtime archives reuse complete cache bytes, freshly rehashed against the new assets and original runtime release with archive-integrity checks. They were not rebuilt or re-executed. SHA256SUMS covers all 12 payloads.
• The 1.4.24 warning still applies: checkpoints trained with both norm_means=True and norm_vars=True need recognition-quality reevaluation. norm_vars=False remains the default.
• No production upgrade or adoption/growth claim.

GitHub Release and official PyPI downloads are both accepted; the plan below is retained history.

Historical Pre-Release Plan

Summary

Release the root FunASR package as 1.4.26:

• fix(server): return plain text for response_format=text #3874: return the bare transcript for HTTP response_format=text, rather than a JSON-quoted string.
• fix(postprocess): keep a space between Latin VAD segments in rich transcription #3872 / fix(postprocess): preserve VAD word boundaries with current-main CI #3883: preserve word separators across English/Korean VAD segments while retaining CJK and event behavior.

The eight release files update the package version, four READMEs, both repository-role guides and their version regression contract. Original release history and the normalization/checkpoint warning remain.

Validation

• Version/docs contract: RED eight failures before the release edits, GREEN 53 passes afterward.
• Other selected release/docs contracts: 102 passes, two browser cases skipped locally and required in hosted CI.
• Actual installed candidate wheel: 111 passes, 34 independent-runtime cases deliberately deselected. No unresolved skips in this installed-root selection.
• Strict Twine, offline installation, import and CLI checks pass. All 439 package source files and 13 entrypoints are checked against the candidate tree.
• fix(postprocess): preserve VAD word boundaries with current-main CI #3883 integration already completed 246 selected CPU cases; fix(server): return plain text for response_format=text #3874 retained 32 actual HTTP/SDK and 213 related CPU cases. Their failures and coverage limits remain preserved.
• Root model/GPU-runtime source is unchanged from accepted 1.4.25 and bound to its existing L20Z evidence. No new GPU execution, full-root-suite, pretrained ASR-quality, Nano or production-upgrade claim.

New exact-head, merged-main and final-tag CI, checkout-tree acceptance, actual GitHub assets and manifest, final installed payload checks and official PyPI downloads are separate gates. PyPI is last.

Independent ONNX/Torch abbreviation and timestamp fixes need their own distribution releases; this root wheel does not publish those packages. Open #3861, #3862, #3867 and new hotword PRs are not claimed resolved.