# Local English speech engine

The standalone worker uses the unmodified **speak.js** engine from
https://github.com/kripken/speak.js at revision
`9c6f642f7d78bab51d16b2e8b79cdf205643ec35` (eSpeak 1.45 source tree).
It is distributed under GNU GPL version 3 or later, without warranty.
See [LICENSE.txt](LICENSE.txt) and [AUTHORS.txt](AUTHORS.txt).

[Download the complete corresponding upstream source](speak-js-9c6f642f7d78bab51d16b2e8b79cdf205643ec35-source.tar.gz).
This same-server archive includes the C++ engine, dictionary and voice data,
`src/Makefile`, the Emscripten build scripts and the original prebuilt
`speakGenerator.js`. It is supplied with the engine at no additional charge.
SHA-256: `af30b74e164bdcaef792185ae1bfbe533c9c7a5226643cda5ed37cee7bd0eed5`.
The unmodified engine JS SHA-256 is
`d83ce9905647988e7496fed48f897564cab264580c77de4d3191cb794dd29898`.

Build guidance is in the archive's `README.markdown` and
`src/emscripten.sh`: install a compatible legacy Emscripten toolchain, adjust its
local path in that script, and run it inside `src/`. The original build script
uses legacy Emscripten/Python interfaces. No claim is made that current
Emscripten releases reproduce that historical build byte for byte.
The archive preserves the complete original build inputs and instructions.
FILKO ships that revision's unmodified prebuilt engine, not a new compilation.

FILKO's small worker adapter is licensed under GPL-3.0-or-later as well.
Its complete readable source is [speech-worker.js](speech-worker.js);
the repository copy is `src/web/workers/speech-worker.js`.
`website/scripts/prepare-speech.mjs` verifies the committed hashes and copies
the adapter unchanged. The engine runs in a separate classic web worker with
text/speed input and WAV-byte output. Files never leave the device.

English voice only; robotic synthesized speech, not a neural voice, translation,
OCR, speech recognition, or MP3 encoding. The engine has no warranty; recipients
may copy, modify and redistribute it under the included GPL terms.
