4042fc53a68f1b24d14dbddd1a09ff8945cee6f0
3
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Bohdan Triapitsyn
|
391f938334 |
feat(voice): match local and macOS voices to the language of the text
Text-to-speech picked one voice regardless of what language a reply was in. A dependency-free language detector (script, marker letters, function words) now decides the language of the whole message once; with the new "Match the voice to the language of the text" setting the local provider switches to a catalog model for that language (Kokoro zh/en and Piper models for 12 languages, downloaded on first use like the existing model) and macOS say switches to an installed voice whose locale matches. The local voice picker lists voices of every installed model, and the settings show which language models are on disk. The Ukrainian Piper medium build is a character-level model that sherpa-onnx turns into noise, so the espeak-based Lada build is used instead. Claude-Session: https://claude.ai/code/session_017TK5JAYDfT3Fotc23UEg98 |
||
|
Serhii DziupinandSerhii Dziupin
|
86e6a2ae76 |
Remove verified dead declarations (#2714)
* chore: remove verified dead declarations Co-authored-by: Serhii Dziupin <makeittech@users.noreply.github.com> * chore: narrow unused internal exports Co-authored-by: Serhii Dziupin <makeittech@users.noreply.github.com> * chore: remove newly exposed dead helpers Co-authored-by: Serhii Dziupin <makeittech@users.noreply.github.com> * chore: remove unused deep-link serializer Co-authored-by: Serhii Dziupin <makeittech@users.noreply.github.com> * test: drop two tests that assert on copies of the code mainLayoutMobileSidebarMount read MainLayout.tsx and SessionSidebar.tsx as strings and asserted on source substrings down to exact indentation, so it failed on formatting rather than behaviour. useProjectSessionSelection.test reimplemented the hook's visitNodes logic inside the test file and asserted against that copy, so it could not observe the hook at all. Co-authored-by: Serhii Dziupin <makeittech@users.noreply.github.com> * test: repair sync suites that had rotted while unrunnable No runner executed packages/ui, so these drifted from the source unnoticed: two imported helpers that are no longer exported, one directory-store stub predated the session field routeMessage reads, and the WebSocket fake missed the mandatory url-token mint plus the close event the socket wrapper reads. Co-authored-by: Serhii Dziupin <makeittech@users.noreply.github.com> * test: stop the web suite failing on timeouts and a hand-copied mock The Git suites drive a real git binary, so the 5s default made a valid suite fail differently per run. The gitApiHttp mock listed ~70 export names by hand and fell behind the source; it now derives every stub from the real module, which the added shared-UI aliases make resolvable. Co-authored-by: Serhii Dziupin <makeittech@users.noreply.github.com> * test: run every suite from one command and in CI packages/ui (232 files) and packages/vscode (22) had no test script at all, CI ran neither, and 9 vscode files could never run because Node cannot resolve their extensionless TypeScript imports. Three electron files sat outside every script list, one of them importing vitest, which that package does not depend on. A runner gives each file its own process, since these suites keep module-level singletons and fail by load order when sharing one. Co-authored-by: Serhii Dziupin <makeittech@users.noreply.github.com> * chore: delete a superseded repro harness and a completed plan The issue-2638 harness needed lsof, overrode process.platform and spawned real servers, and nothing referenced it; event-stream/rebind.test.js now covers the same hub-pinned-to-the-old-port behaviour. The pairing v2 plan described relay and the pairing UI as out of scope, both of which shipped. Co-authored-by: Serhii Dziupin <makeittech@users.noreply.github.com> * docs: point at the theme tools and record the github barrel invariant convert-vscode-theme and harmonize-theme were referenced nowhere, so the theme-authoring reference now names them. The github barrel is loaded through await import('./index.js') and destructured per route, which no static report can see; documenting that is what stops the next cleanup from deleting it. Co-authored-by: Serhii Dziupin <makeittech@users.noreply.github.com> * test: repair merge drift in bridge and route-registry mocks upstream/main gained upsertProviderConfig on bridge-system-runtime and a PATCH scheduled-task route after this branch forked. Their test doubles were never updated to match: - bridge-system-runtime.test.js: add upsertProviderConfig to the opencodeConfig mock so the import resolves. - sse-routes.test.js: add app.patch to the route registry stub. --------- Co-authored-by: Serhii Dziupin <makeittech@users.noreply.github.com> |
||
|
Bohdan Triapitsyn
|
de1b85ac56 |
feat(voice): first-class voice input and local TTS across web, desktop, and mobile (#2018)
Complete rebuild of voice input on a server-authoritative streaming architecture, replacing the legacy Web Speech / whole-blob / WASM engines and the dead voice-agent layer (~4k lines removed). Speech-to-text (dictation): - Client streams 16 kHz mono PCM16 chunks over /api/dictation/ws with seq/ack ordering; buffered audio is retained and replayed on reconnect - Server transcribes and streams live partial transcripts back; segments auto-commit every ~15s with silence suppression and adaptive finalization timeouts - Local provider (default, zero config): sherpa-onnx models in a forked worker process — auto-download with progress, staged extraction with verification, corrupt-model auto-recovery, idle shutdown after 5 min - Model catalog with settings picker (accuracy/speed ratings, sizes, download/delete): Parakeet TDT v2 (English) and v3 (25 European languages, auto-detected), Whisper base and tiny (multilingual, light) - OpenAI-compatible provider for any Whisper endpoint - Composer overlay with live transcript, volume meter, timer, and cancel / insert / insert-and-send actions; failed transcriptions keep their audio for retry or accepting the partial text as-is - Configurable keyboard shortcut (default mod+alt+v) toggles dictation; Enter confirms and Escape cancels while recording - Overlay is pixel-aligned with the composer (measured footer height, matching paddings/typography/gaps) — no layout shift when toggling Text-to-speech: - Local Kokoro provider (English, 11 voices) synthesized in the same worker via /api/dictation/tts/speak, managed by the shared model pipeline; sentence-pipelined playback keeps time-to-first-audio at ~1 sentence regardless of message length, and stop cancels in-flight synthesis - Sanitizer keeps inline-code content (strips backticks only), reads interword slashes aloud, and removes only absolute file paths Settings: - Voice page unified: a single read-aloud toggle owns all playback options (the confusing "Enable Voice Mode" is gone); a new "Enable voice input" toggle (default on, persisted to settings.json) hides the composer mic entirely when disabled Mobile and transport: - iOS/Android microphone permissions added (dictation was previously impossible on mobile) - Fixed Android WebSocket upgrades: the Capacitor WebView origin (https://localhost) was missing from the packaged-client allowlist, 403-ing every WS connection — root cause of the old mobile SSE lock, which is now removed for all transports Security and conventions: - All HTTP routes sit behind the global /api auth gate; the WS upgrade explicitly validates the UI session and origin, with oc_url_token narrowly allowlisted and covered by tests; the dictation socket mints a fresh URL token before connecting - Routes register before the generic OpenCode proxy; the client goes through runtimeFetch/getRuntimeUrlResolver, and runtime switches reset the dictation socket - VS Code deliberately reports dictation as unavailable (no server process in that runtime) CI: workflow Node bumped 20 -> 22 to match the repo engines and fix better-sqlite3 installs broken by node-gyp@latest on Node 20. New dependency: sherpa-onnx-node (prebuilt N-API; macOS/Linux x64+arm64, Windows x64 — Windows-on-ARM falls back to the OpenAI-compatible provider) |