Commit Graph
5 Commits
Author SHA1 Message Date
eab3d45192 Fix runaway MLX Qwen audio chunks (#964)
* fix runaway MLX Qwen audio chunks

* test: tighten runaway retry coverage

---------

Co-authored-by: huanghua01 <[email protected]>
2026-07-26 23:33:09 -07:00
Jamie Pine b434db22f6 chore(backend): repair test suite and bring ruff to green
The suite hadn't run green since the routes refactor:
- test_profile_duplicate_names.py imported the pre-refactor module
  layout and broke collection; now imports backend.services.profiles
- tests/conftest.py puts the repo root and backend dir on sys.path so
  files collect standalone instead of depending on run order
- test_cors.py tested a hand-copied mirror of the origin list that had
  drifted from app.py (missing http://tauri.localhost); it now builds
  the app via the real create_app() factory
- test_progress.py simulated a 1KB download, below the tracker's 1MB
  reporting threshold; simulation raised to 5MB
- slow/timeout markers registered in pyproject

Ruff: ~900 violations auto-fixed (typing modernization, import
sorting, unused imports, whitespace). The remaining rules are baselined
in pyproject.toml with per-rule counts to burn down, plus per-file
carve-outs for deliberate env-before-import ordering. ruff check is
now clean; suite is 134 passed, 2 skipped.
2026-07-26 23:16:09 -07:00
James Pine b7781951df comment cleanup 2026-03-16 01:46:19 -07:00
James Pine 97292ecef7 feat: add chunk crossfade slider (0ms = hard cut)
Persisted setting (default 50ms) controls how audio chunks are blended
together.  Set to 0 for a clean hard cut with no overlap.
2026-03-13 06:48:06 -07:00
James Pine 70ca7f66cb feat: chunked TTS generation for long text (engine-agnostic)
Text exceeding max_chunk_chars (default 800) is automatically split at
sentence boundaries, generated per-chunk, and concatenated with a 50ms
crossfade.  Works with all engines (Qwen, LuxTTS, Chatterbox, Turbo).

- Abbreviation-aware sentence splitter (Dr., Mr., e.g., decimals)
- CJK sentence-ending punctuation support
- Paralinguistic tag preservation ([laugh], [cough], etc.)
- Per-chunk seed variation to avoid correlated RNG artefacts
- Per-chunk Chatterbox trim (catches hallucination at each boundary)
- max_chunk_chars exposed as per-request param on GenerationRequest
- Text max_length raised to 50,000 characters

Closes #99
2026-03-13 06:21:34 -07:00