Skip to content

V2 - #1

Merged
temdah merged 20 commits into
masterfrom
v2
Aug 3, 2026
Merged

V2#1
temdah merged 20 commits into
masterfrom
v2

Conversation

@temdah

@temdah temdah commented Aug 3, 2026

Copy link
Copy Markdown
Owner

No description provided.

temdah added 20 commits August 2, 2026 18:01
Persist deterministic, size-bounded sentence anchors and expose dedicated audio and timing cache identities. Preserve sentence caches across rechunking while isolating changes to voice, language, model, and synthesis steps.
Route sentence-aligned starts through stable low-latency cache units and prefetch subsequent speech for continuous playback. Preserve exact mid-sentence behavior through the existing fallback while exposing sentence cache hits and misses in diagnostics.
Serialize ONNX inference and let foreground playback take priority over prefetching and audiobook rendering. Invalidate stale requests and rebuild failed cached audio once before presenting a manual retry.
- Plan punctuation-aware cadence and stream trailing silence through native AAC encoding without duplicating waveforms
- Version affected caches, expose dev diagnostics, and shorten playback and shared-type imports through their barrels
- Add configurable semantic tone planning with natural pacing and pause profiles.
- Reduce lower-end device pressure with adaptive prefetching, O(1) queueing, and a five-step synthesis floor.
- Replace cross-module subpath imports with the shortest public barrel paths.
- Export shared engine, PDF, and utility helpers required by consumers and tests.
- Move playback-owned hooks into the playback module
- Add an internal core barrel while preserving the short public playback API
- Remove the playback-to-hooks-to-playback dependency cycle
- Run barrel-based tests through the Expo-aware Jest environment
- Isolate native ONNX, audio, codec, and device modules during unit tests
- Require all unit tests to pass before CI typechecks and builds the APK
Serialize engine loading, swapping, and release around active inference. Honor low-memory warmup behavior and prevent cache metadata failures from interrupting playback.
- Move voice resolution and clip synthesis behind a profile-scoped playback service
- Preserve request cancellation and engine priority with focused unit coverage
- Move audio mode, lock-screen metadata, artwork, and rate synchronization into a dedicated playback hook
- Centralize notification teardown and cover metadata and rate policies with unit tests
- Keep usePlayback as a small facade over preparation, player control, adaptive prefetch, recovery, media-session, and position modules
- Add focused coverage for cache decisions, sequencing, retries, seeking, and document position mapping
- Move reader rendering, overlays, and developer actions into reusable components
- Centralize benchmark orchestration and document deletion behind shared services
- Move shared contracts into cohesive type files and expose them through public barrels.
- Remove redundant comments while preserving explanations for playback, caching, algorithms, and platform constraints.
@temdah
temdah merged commit 122a7ed into master Aug 3, 2026
1 check 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.

1 participant