Skip to content

Add speech voice/generate and Qwen-MT translate commands - #329

Open
hopelynd wants to merge 4 commits into
mainfrom
feat/speech-voice-and-translate
Open

hopelynd wants to merge 4 commits into
mainfrom
feat/speech-voice-and-translate

Conversation

@hopelynd

@hopelynd hopelynd commented Nov 20, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

  • Add speech voice enrollment/management (clone/design/list/get/update/delete/preview) and speech generate, wired through core enrollment client + bl command map
  • Add Qwen-MT translate surfaces: text translate, image translate, and omni translate, with dry-run / async wait and skill + reference updates
  • Share Chat Completions streaming via run-chat (snapshot vs delta SSE for qwen-mt-plus) and extend task-get projections used by image translate

Description

Speech voice + generate

  • New commands under speech voice * for clone / design / list / get / update / delete / preview
  • New speech generate for text_prompt / references driven audio generation (WAV / MP3 / OGG Opus)
  • Core: enrollment client, types, endpoints; dry-run redaction; save-error voice_id hints; out-prefix checks
  • Tests: offline e2e plus gated live CRUD; bailian-gen skill / reference/speech.md updated

Qwen-MT translate

  • bl text translate — dedicated text MT (default flash family); empty input rejected locally
  • bl image translate — async image MT with wait / task-get projection for image_url and Success/Data shapes
  • bl omni translate — omni-side MT path aligned with shared streaming helpers
  • Chat Completions extracted to run-chat; stream parser supports snapshot vs incremental (delta) SSE for qwen-mt-plus / flash-lite
  • Skills: hub keeps named Bailian / Qwen-MT / bl text translate; image·doc·PDF translation hand-off stays on bailian-gen

- Add speech voice clone/list/get/update/delete/preview and design
- Add speech generate (text_prompt/references; WAV/MP3/OGG Opus only)
- Wire core enrollment client, types, endpoints, and bl command map
- Harden dry-run redact, save-error voice_id hints, and out-prefix checks
- Cover with offline e2e plus gated live CRUD; update bailian-gen skill
… streaming

- Add `text translate`, `image translate`, and `omni translate` with dry-run, async wait, and skill/reference updates
- Extract Chat Completions into `run-chat` and support snapshot vs delta SSE for qwen-mt-plus
- Extend task-get projections (`image_url`, Success/Data) and reject empty MT messages locally
- Cover empty input, flash/lite incremental stream, plus text-mode suffixes, and image CANCELED/UNKNOWN offline

This branch has not been deployed

No deployments
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.

1 participant