Skip to content

fix: support sync-flash and qwen3-filetrans ASR models in speech recognize - #149

Merged
XXPermanentXX merged 6 commits into
mainfrom
fix/fixed_issue_146
Aug 13, 2026
Merged

fix: support sync-flash and qwen3-filetrans ASR models in speech recognize#149
XXPermanentXX merged 6 commits into
mainfrom
fix/fixed_issue_146

Conversation

@hopelynd

Copy link
Copy Markdown
Collaborator

Summary

  • Add asr-routes.ts with resolveAsrApi() to dispatch ASR models to the correct DashScope endpoint; previously all models were routed to the async filetrans path, causing url error, please check url! for sync-flash models (fixes bl speech recognize 提示 url error, please check url! #146)
  • Sync-flash models (fun-asr-flash*, qwen-audio-*-asr-flash, qwen3-asr-flash*) are now routed to multimodal-generation with the correct request body per flash family (qwen3 vs input-audio)
  • qwen3-asr-flash-filetrans* now correctly uses file_url (singular) instead of file_urls; realtime/streaming models surface a clear USAGE error; routing logic is shared between the CLI command and the pipeline step

…gnize

- Add asr-routes.ts with resolveAsrApi() to route models to the correct
  DashScope endpoint instead of always hitting asr/transcription
- Async filetrans: fun-asr / paraformer / *-filetrans → file_urls (plural)
- Async filetrans (qwen3): qwen3-asr-flash-filetrans* → file_url (singular)
- Sync flash (input-audio): fun-asr-flash* / qwen-audio-*-asr-flash → multimodal-generation
- Sync flash (qwen3): qwen3-asr-flash* → multimodal-generation + asr_options
- Realtime/streaming models now give a clear USAGE error instead of a
  confusing server-side "url error"
- Propagate same routing logic to pipeline speechRecognize step
- Add table-driven unit tests and dry-run e2e assertions
Fixes #146
@hopelynd
hopelynd requested a review from XXPermanentXX August 12, 2026 09:29
Add SSE disable header, data-URI format inference, broader response text
parsing, HTTP contract e2e, pipeline routing tests, and ASR model selection
guidance in bailian-gen.
…d_issue_146

# Conflicts:
#	skills/bailian-gen/reference/index.md
#	skills/bailian-gen/reference/speech.md
Normalize async ASR transcription items so waiting mode downloads text and --out works without changing shared media task types.
@XXPermanentXX
XXPermanentXX merged commit daefc09 into main Aug 13, 2026
2 checks passed
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.

bl speech recognize 提示 url error, please check url!

2 participants