Repository navigation
fix(postprocess): preserve VAD word boundaries with current-main CI - #3883
Merged
Merged
Conversation
…nscription AutoModel joins the tagged SenseVoice text of each VAD segment with a single space, so a long English recording reaches rich_transcription_postprocess() as several <|lang|> blocks. The function strips every block and concatenates them with no separator, so "Hello world." and "How are you?" become "Hello world.How are you?" and the last word of one segment runs into the first word of the next. Insert a space between two blocks when both boundary characters belong to a space-separated script (ASCII or Hangul). Chinese and Japanese text and emoji boundaries are still joined without a space. The same change is applied to the funasr_onnx and funasr_torch copies so the three shipped postprocessors stay identical, and the event-segment test that pinned the glued "helloworld" output now expects "hello world". Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Preserve the original PR #3872 author commit. Synchronize current main content so the exact-head text workflow includes test_type_utils_compat.py. No test removal, history rewrite, or protection bypass. Signed-off-by: LauraGPT <zhifu.gzf@alibaba-inc.com>
…esponses Correct the Product site CI failure exposed by #3874 without changing the handler or weakening security/format contracts. Preserve the 1.4.15 recipe and add visible, source-bound current-main notices. Signed-off-by: LauraGPT <zhifu.gzf@alibaba-inc.com>
Complete the prior four-article edit with its strict manifest and per-article modification date. Preserve all other snapshot entries, original capture date, historical recipe and exact hash assertions. Signed-off-by: LauraGPT <zhifu.gzf@alibaba-inc.com>
Merged
LauraGPT
added a commit
that referenced
this pull request
Nov 12, 2026
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
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.
Complete site snapshot validation
The second companion head fixed the old HTTP assertion but exposed five integration failures: four source/built article-date checks and the strict legacy hash manifest. The date fixture now records the actual changed article date, and only the four reviewed article hashes are updated. Historical capture date and all remaining manifest entries stay unchanged; no check is removed or relaxed.
The exact five failures reproduced locally (16 other metadata cases passed). The complete product-site Python suite now passes all 532 cases with no skips. Browser and other new-head hosted checks remain separate gates. Prior failures are retained; production website files and services remain unchanged.
Site CI follow-up
The first companion head exposed a stale Product site contract: four historical articles pin FunASR 1.4.15, but the test incorrectly required current main to keep the old JSON-quoted text response. #3874 already changed main to a plain-text response.
The follow-up retains the old recipe, adds visible version-boundary notices in all four English/Chinese source and built pages, and checks the current handler separately. Eight missing-notice regressions fail before the documentation change; 69 targeted source/built contracts pass afterward. The earlier site run (one failure, 523 passes) is retained. No production deployment or handler behavior change is made here.
New-head CI must be accepted before merge; earlier passed checks are not substituted for it.
Summary
Integrates #3872 while retaining its original commit and authorship. English/Korean VAD blocks keep their word separator; CJK and event/emotion behavior remains covered.
The original exact-head text CI could not collect tests: its workflow referenced
tests/test_type_utils_compat.py, which was added on newer main and absent from the old head. The signed companion commit retains current main content instead of removing the test or bypassing the failure.Validation