v0.2.1

Fixed

• setVoice (and the voice option of synthesize and phonemes) now
  accepts a language tag that has no voice file of the same name, such as
  "en-gb" or "en-gb+f3". A string that matches no voice name is retried
  as a language when a voice lists exactly that language tag (ignoring case),
  and throws otherwise.
• Loading data over HTTP no longer fails when the server sends the .gz
  files with Content-Encoding: gzip, as Vite's dev server does for files
  imported with ?url. The browser has already unzipped those, so
  httpDataSource now passes on a file that does not start with the gzip
  magic bytes as it is.
• A data file that fails to decode now names the file in the error.

v0.2.0

espeak-ng 1.52.0, unchanged.

Added

• @spelling-creator/spellophone/wasm2js: espeak-ng compiled to plain
  JavaScript by Binaryen's wasm2js, for engines without WebAssembly
  (createWasm2jsModule).
• A fallback option for the browser createEspeak. Pass
  () => import("@spelling-creator/spellophone/wasm2js") and the wasm2js
  module is used where WebAssembly is blocked, such as Safari in Lockdown Mode
  or under a Content-Security-Policy without wasm-unsafe-eval.
  canCompileWasm() exposes the check it makes.
• A legacy build lowered to ES5 with swc, with the polyfills it needs and
  no WebAssembly: @spelling-creator/spellophone/legacy as an ES module, and
  dist/legacy/spellophone.iife.js as a classic script that defines a
  Spellophone global.
• httpDataSource takes an optional third argument that replaces the gzip
  decoding.

Changed

• The package is now published to npmjs.com instead of GitHub Packages, so
  installing it no longer needs a GitHub token or an .npmrc entry. The
  tarball is still attached to each GitHub release. Versions after this one
  also carry a provenance attestation.
• cdn: "jsdelivr" and jsdelivrDataUrl now use jsDelivr's npm endpoint
  (cdn.jsdelivr.net/npm/@spelling-creator/spellophone@<version>/) instead
  of its GitHub endpoint.
• EspeakModule (and what DataLoader and Espeak.create accept) no longer
  includes NODEFS, which the wasm2js module does not have. MainModule from
  @spelling-creator/spellophone/wasm/espeak-ng.js is still the full type.
• The docs site uses the Spelling Creator favicon, and its header links to
  the package on npmjs.com.

Development

• The compiled wasm and data are no longer committed, and were removed from
  the repository's history. pnpm build compiles them from espeak-ng's
  sources, and CI builds them on Linux and macOS and fails if the two builds
  differ in any byte.
• The release workflow publishes with npm's trusted publishing (OIDC), so the
  repository holds no npm token.
• The tsdown config uses deps.neverBundle in place of external, which
  tsdown 0.23 deprecates. The output is the same.

v0.1.0

First release. espeak-ng 1.52.0.

Added

• createEspeak for Node.js (@spelling-creator/spellophone/node) and for
  browsers and workers (@spelling-creator/spellophone/browser); the bare
  import picks the right one.
• Espeak.synthesize: 16-bit PCM audio with word, sentence, end, mark, play
  and phoneme events; per-call rate, pitch, range, volume and
  wordGap; SSML and [[ ]] phoneme input.
• Espeak.phonemes: IPA or espeak-ng ASCII transcription with optional
  separator and ties.
• Voice listing and selection by name, language or properties, including
  espeak-ng's +variant syntax.
• Dictionaries load on demand: synchronously from disk in Node.js, over HTTP
  in browsers with languages, loadLanguages and loadVoice.
• Data served from your own host (dataUrl) or from jsDelivr's GitHub
  endpoint (cdn: "jsdelivr"); all data files gzipped and decoded with
  DecompressionStream.
• encodeWav and toFloat32 helpers.
• Low-level Espeak.create, DataLoader, DataSource, httpDataSource
  and fileDataSource for custom module and data loading.