- Remove hardcoded preferred_provider=huggingface in podcast handlers
- Set preferred_provider=None to respect GPT_PROVIDER env var
- Change default model from Qwen to gpt-oss-120b:cerebras (the model user had access to)
- WaveSpeed will now use gpt-oss-120b model instead of Qwen
- Add dedicated image_generation module with statistical extraction
- Support 16 industry domains with visual concept detection
- Add model-specific guidance for Ideogram, FLUX, GLM, Qwen, MAI
- Extract statistics, rankings, comparisons, and trends automatically
- Refactor backend/api/images.py to use new module
This commit adds the Auto-Dubbing feature for Podcast Maker with support
for translating podcast audio to different languages with optional voice
cloning to preserve the original speaker's voice.
New Features:
- Translation Service (common module): DeepL integration for low-cost
translation, WaveSpeed integration for high-quality translation
- Audio Dubbing Service: STT -> Translate -> TTS pipeline with
voice cloning support
- 9 new API endpoints for dubbing and voice cloning
- Support for 34+ languages
- Cost estimation utilities
- Comprehensive documentation
Files Added:
- services/translation/ (5 files): Translation service module
- services/dubbing/: Audio dubbing service
- api/podcast/handlers/dubbing.py: API endpoints
- docs/AUTO_DUBBING.md: Feature documentation
- CHANGELOG.md: Change log
Files Modified:
- api/podcast/models.py: Added dubbing request/response models
- api/podcast/router.py: Added dubbing routes
- services/__init__.py: Export translation and dubbing services
- scene_animation.py: Fixed missing Path import
- Add get_current_user authentication to all user data endpoints
- Pass authenticated user_id from auth context to service methods
- Add proper HTTPException handling for missing data
- Fix user_id type from int to str in service methods
- Ensure endpoints only return data for authenticated user
- Modified _ensure_initialized() to run in background thread (non-blocking)
- Added _ensure_initialized_async() for truly async initialization
- Updated index_content() to return immediately without waiting for initialization
- Weights now load in background thread instead of blocking event loop
- Added initialization tracking to prevent duplicate initialization
- Modified today_workflow API to handle non-blocking indexing gracefully
- This prevents dashboard refresh from blocking other services
When a user accesses the dashboard, the indexing now happens in background
instead of blocking the HTTP response, allowing other services to function
normally while weights are being loaded.
- Use canonical Clerk user id (clerk_user_id) across all onboarding entrypoints to ensure consistent OnboardingSession.user_id lookup
- Fix API key persistence in api_key_manager.py to use correct APIKey model columns (session_id, provider, key)
- Increase Node heap for frontend build to 8GB and add build:nomap script to disable sourcemaps and reduce memory usage
- Update onboarding endpoints (endpoints_core.py, onboarding_control_service.py, step_management_service.py) to prefer clerk_user_id over id
- Fix frontend workflowStore.ts TypeScript error by returning WorkflowError instance
- Add website_automation_service.py for onboarding automation
- Add TaskStatusEnum to enumerate valid status values (pending, in_progress, completed, skipped, dismissed)
- Add TaskStatusUpdateRequest Pydantic model with validation
- Constrain completion_notes to max 4000 characters
- Automatically enforce schema validation and improve OpenAPI docs
- Update set_task_status endpoint to use typed request body
- Remove need for manual status validation (FastAPI handles it)
- Preserve dependencies normalization helper and all usages
- Preserve date validation and narrower exception handling from PR #396
- Keep proper feedback scoring using task.status from database
- Keep contextuality validation response fields intact
- Maintain all observability and error handling improvements
- Improve API robustness through type safety
- Add _normalize_dependencies() helper to handle all dependency type variations
- Handle None, list, JSON string, and invalid types with safe fallback to []
- Apply normalization to today and yesterday task payloads for consistency
- Ensure indexing pipeline receives normalized list dependencies
- Preserve task status feedback scoring logic (uses task.status, handles all negative cases)
- Keep contextuality validation and quality status response fields
- Improve data consistency across API and indexing surfaces
