⚠️This is a Work in Progress

saymore-plugin

A SayMore-compatible audio annotation SPA — Manual Segmenter, Auto Segmenter, and

the Transcription / Free-Translation grid — to be packaged as a lameta

file-handler plugin. Originally developed as a sub-package of the lameta repo,

now its own repository.

The authoritative spec (architecture, module layout, interface signatures,

milestones, risks, execution status) lives in docs/PLAN.md.

This README covers only how to run and test the package.

Ground rules

• Vite+ (vp) + pnpm — never npm or yarn. Run vp install after pulling. See

  AGENTS.md.

• Own package.json + pnpm-lock.yaml. pnpm-workspace.yaml exists only to hold

  the Vite+ catalog that pins tooling versions; there are no sub-packages.

• Own React 18 (the plugin hosts this in an iframe/webview), but keeps lameta team

  conventions: Emotion css prop, MobX 6 + observer, colocated *.spec.ts.

• Built on Vite+ (Vite 8 / Rolldown / Vitest 4 / Oxlint / Oxfmt) with the TS 6

  preview compiler.

Scripts

vp install    # install deps (pnpm under the hood)
vp dev        # Vite dev server — renders the shell
vp build      # production build (Vite + Rolldown) → dist/
vp preview    # serve the production build locally
vp test       # Vitest (node env by default; component specs opt into happy-dom)
vp check      # format (Oxfmt) + lint & type-check (Oxlint) — run before committing

Layout

src/
  main.tsx, App.tsx    # shell: plugin-iframe or OpenScreen → Manual Segmenter
  plugin/              # lameta plugin client kit + PluginHostAdapter + manifest glue
  l10n.ts              # t(id, default) seam; no Lingui yet
  model/               # TimeRange, AnnotationSegment, EAF doc, BoundaryRules, ...
  audio/               # envelope, EnvelopeCache, PlaybackEngine, autoSegmenter
  fs/                  # FileSystemAdapter seam + InMemory / BrowserDirectory adapters
  state/               # ProjectStore, AnnotationDocumentStore, view models
  components/          # waveform surface + overlay, segmenter, grid, shell
  harness/             # dev-only debug panels
test-data/             # fixtures — see test-data/README.md

Status

Working Manual Segmenter vertical slice + auto-segmenter core. In the dev

harness (vp dev) you can drop a WAV or open a SayMore session folder (Chromium

only — File System Access API), see the waveform, segment with SayMore parity, and

write a valid .annotations.eaf. The Auto Segmenter has a faithful algorithm port

(no dialog yet); the Transcription grid is not built yet. The package can also be

built and run inside lameta as a file-handler plugin (see below). See

docs/PLAN.md for the full milestone breakdown and remaining phases.

Packaging as a lameta plugin

The SPA doubles as a lameta file-handler plugin: when embedded in lameta's plugin

iframe it connects over postMessage, wraps the host file API in a

PluginHostAdapter (src/plugin/), and opens the selected file's session — the

same ProjectStore the dev harness uses. public/plugin.json5 is the manifest

(one Audio tab, companionFiles permission); Vite copies it to the build root,

so dist/ is itself a ready-to-use plugin folder.

vp build              # → dist/ (index.html + assets + plugin.json5)
vp run package        # build, then zip dist/ → saymore-audio.lmplug

Live dev loop: run vp build --watch, then in lameta turn on Developer mode →

File → Plugins… → Developer plugin folder and point it at this repo's dist/.

lameta reloads the open tab whenever the build re-emits. Install a finished

.lmplug via File → Plugins… → Install plugin….

Manual test protocol

(To be filled in as tools land — Phase 4.) The compatibility gate is: copy a real

SayMore session → edit in this SPA → reopen in SayMore and ELAN, verify

playback and that foreign tiers survive.