# Turkish eSpeak-NG source and licenses

The standalone voice engine uses the unchanged prebuilt files from
`echogarden-project/espeak-ng-emscripten` commit
`7ab07eba2d966ce45040c88d9be953e1d68640e7`. Its `package.json` declares
`@echogarden/espeak-ng-emscripten` version **0.4.1**, GPL-3.0. These are direct
GitHub commit artifacts, **not an npm 0.4.1 tarball**: that version was not
published in the npm registry when this package was prepared.

| Upstream file | Bytes | SHA-256 | Verified Git blob SHA-1 |
| --- | ---: | --- | --- |
| `espeak-ng.js` | 603777 | `ba2b5f80f00a782a467c2072df2afe277124e328083ef6fb1aeed7d40205342a` | `c28eaf688ac3933f0752cabb524918bd3e21b415` |
| `espeak-ng.data` | 19099765 | `34d8d90d112acd35f6b7cdf3da16125444aeeebc4b68bee87cd55d6f7e4dd3a0` | `d825e74b9d950f0098ddf708f4be7545a310dc96` |

Both files were independently checked inside the pinned source archive. The
JavaScript contains embedded WASM; its companion data includes language data.
The source fork explicitly contains the Turkish `tr` voice. This is formant
synthesis, with robotic timbre, rather than a natural neural voice.

## Source archives

The following archives were downloaded through verified TLS and copied without
repacking. Their complete source trees retain upstream copyright and license
notices. Neither source tree has Git submodules.

| Archive | Bytes | SHA-256 |
| --- | ---: | --- |
| [Wrapper source](sources/espeak-ng-emscripten-7ab07eba2d966ce45040c88d9be953e1d68640e7.tar.gz) | 9827053 | `bb66e1e44fb8ff03577a26e627d872a9efce6734d05b0ada9617d9bead4fc45f` |
| [Engine and build source](sources/espeak-ng-9a550bef455f03b459f51796e3482833aab7fbc0.tar.gz) | 18689115 | `06afb088f1e9b0546c8be0c0c3295ad7851a846a1242e9850856dbfb6a344271` |

Official archive URLs:

- https://codeload.github.com/echogarden-project/espeak-ng-emscripten/tar.gz/7ab07eba2d966ce45040c88d9be953e1d68640e7
- https://codeload.github.com/echogarden-project/espeak-ng/tar.gz/9a550bef455f03b459f51796e3482833aab7fbc0

## Rebuild instructions and limits

Extract the engine archive into a separate build directory outside the app
checkout. On Linux, provide CMake, a C/C++ compiler, Make, Python and an activated
Emscripten SDK. Record and pin their versions: upstream does not provide a
complete toolchain lock for this prebuilt distribution. From the extracted
engine directory, run its unchanged `build-emscripten.sh`.

That script builds native language data, then a static Emscripten engine and
`emscripten/espeak-ng.js` plus `emscripten/espeak-ng.data`. It disables MBROLA,
asynchronous synthesis, libpcaudio and libsonic; the Emscripten build enables
SpeechPlayer and Klatt. Its Makefile explicitly enables WASM, embeds it in the
JavaScript, and exports an ES module. The unchanged build script has SHA-256
`efaa7a8866d924a53a5d73225e672c49e04d5f6c7cfa2fe2199436da791bc234`.

The wrapper README identifies the companion fork branch; the engine archive
pins that branch to the revision above. These archives freeze the supplied
source and scripts. An exact prebuilt-to-toolchain mapping and a bit-identical
rebuild have not been demonstrated. Source provenance alone does not establish browser compatibility or quality.
The companion README records an actual Node/Chromium worker PCM proof; Turkish
pronunciation quality, other devices and production acceptance remain separate.

## Licenses

[LICENSE.txt](LICENSE.txt) is the full unchanged GPL version 3 text. The wrapper
package declares GPL-3.0; eSpeak-NG source states GPL version 3 or later. Commercial
use is permitted subject to the applicable license obligations. Preserve the
notices and supply corresponding source and build instructions when distributing
the GPL engine; review the licensing of any combined distribution separately.

The source archive also retains these separately licensed components. Exact
notice copies are available alongside the archives:

- [Apache 2.0](sources/COPYING.APACHE.txt)
- [BSD 2-clause](sources/COPYING.BSD2.txt)
- [Unicode data terms](sources/COPYING.UCD.txt)
- [ucd-tools GPL text](sources/src-ucd-tools-COPYING.txt)

No Piper neural checkpoints or third-party MBROLA voice models are included.
See [source credits](source-credits.txt) for upstream attribution.