Spaces:
Sleeping
Sleeping
Commit History
-
Fix WS recv race + session recycle 20 with async overlap + Modal deploy
- Fix control_task lifecycle: properly await/cancel before returning to outer WS loop, preventing 'cannot call recv' race condition - Session recycling every 20 sentences (was 200) with async overlap: new ONNX session pre-built in background thread while current session finishes its prefetch queue - Add modal_app.py for Modal serverless deployment - Add soundfile dep for FLAC encoding support91dd53b -
Fix WS recv race + session recycle 20 with async overlap
- Fix control_task lifecycle: properly await/cancel before returning to outer WS loop, preventing 'cannot call recv' race condition - Session recycling every 20 sentences (was 200) with async overlap: new ONNX session pre-built in background thread while current session finishes its prefetch queue - Add soundfile dep for FLAC encoding support8a22550
-
Sync backend with LN-TTS: fix Chapter 1 scraper bug, add sentence offsets, improve error handling, update README
ccb0e29
-
Sentence-chunked WS audio with smoothing
0abd888
-
Fix audio jitter: remove real-time sleep, add ms_start to sentence events
- Remove per-frame asyncio.sleep pacing. Frames now send as fast as synthesis allows, letting the client buffer audio ahead of playback. - Track cumulative_samples per chapter; include ms_start (ms from chapter start) in every sentence JSON event so the client can fire highlights at the correct playback position via getStreamTimeConsumed. - Default prefetch raised to 6 in client (was 3). Co-authored-by: Cursor <cursoragent@cursor.com>8eeaf9e -
Improve chapter parsing, add sentence_total and voice to chapter_info
- scraper.py: more robust chapter number parsing from title and URL (handles Ch., C., Prologue edge-cases, URL fallback) Fixed sort key to preserve stable ordering for unnumbered chapters - server.py: include sentence_total in chapter_info WS message for accurate download progress; include voice field in chapter_info - README.md: add protocol note for sentence_total and chapter parsing Co-authored-by: Cursor <cursoragent@cursor.com>7348977 -
Fix CMD: replace complex inline shell with start.sh script
Co-authored-by: Cursor <cursoragent@cursor.com>4cb1a72 -
Add .gitignore to exclude pycache, models, venv
Co-authored-by: Cursor <cursoragent@cursor.com>923e227
-
Sync backend features: novel_details, realtime flag, portable PORT, Azure deploy docs
- scraper.py: add scrape_novel_details() returning title + cover_url - server.py: add GET /novel_details endpoint; surface in homepage + /info - server.py: WS play accepts realtime=false to skip frame-pacing sleep (fast downloads) - Dockerfile: bind to $PORT (default 7860) for Azure portability - README.md: full Azure Container Apps deploy guide + updated endpoint list Co-authored-by: Cursor <cursoragent@cursor.com>c37e1f4
-
Add Space homepage and runtime info endpoints
Co-authored-by: Cursor <cursoragent@cursor.com>6497be7 -
Deploy existing backend in Docker Space on port 7860
Co-authored-by: Cursor <cursoragent@cursor.com>f8a7e1d -
initial commit
Verified signaturebbe8cec