Hangry LabsUseful tools, easy to run

Third-Party Notices

KokoroTTS includes, adapts, downloads, or depends on third-party software and model assets. Each component remains subject to its own license. This file is informational and does not replace or modify those licenses.

The complete Apache License 2.0 text for this project and the Apache-licensed components below is available in LICENSE.

Kokoro Runtime and Language Processing

Kokoro

  • Project: hexgrad/kokoro
  • Copyright: the Kokoro authors and contributors
  • License: Apache License 2.0
  • Use here: Kokoro model and inference implementation on which this fork is based. Hangry Labs has modified and extended the original implementation for Docker deployment, its standalone UI, APIs, additional languages, audio formats, runtime management, and related tooling.

Misaki

  • Project: hexgrad/misaki
  • Copyright: the Misaki authors and contributors
  • License: Apache License 2.0
  • Use here: grapheme-to-phoneme processing and language support.

phonemizer-fork

  • Project: bootphon/phonemizer
  • Package: phonemizer-fork on PyPI
  • Version currently used: 3.3.2
  • License: GNU General Public License v3 or later
  • Use here: Python interface to eSpeak NG, reached through Misaki for the English out-of-dictionary fallback and for the eSpeak-backed German, Spanish, French, Hindi, Italian, and Portuguese frontends.

The installed wheel includes the GNU GPL v3 license text in its distribution metadata.

These phonemization components are server-runtime dependencies. They are installed by the Docker images and the kokorotts[server] extra; the default dependency-free kokorotts HTTP client installation does not install them.

espeakng-loader and bundled eSpeak NG

  • Project: thewh1teagle/espeakng-loader
  • Version currently used: 0.2.4
  • Loader source license: MIT
  • Bundled component: eSpeak NG 1.52.0 shared library and data
  • Bundled component license: GNU General Public License v3 or later, with additional notices for Unicode data
  • Use here: supplies the eSpeak NG library and data selected by Misaki’s phonemizer integration.

The pinned Misaki runtime explicitly configures phonemizer-fork to use the library and data paths returned by espeakng-loader. This bundled copy is therefore an active runtime dependency, not an unused installation. Matching eSpeak NG source and license material are available from the upstream eSpeak NG 1.52.0 source tree, including COPYING and COPYING.UCD.

At the time this notice was reviewed, the espeakng-loader 0.2.4 wheel did not declare a license in its package metadata or include the relevant license files. Upstream correction is being tracked in espeakng-loader pull request #9. This notice records the licenses and exact corresponding source independently of that packaging omission.

num2words

  • Project: savoirfairelinux/num2words
  • Version currently used: 0.5.14
  • License: GNU Lesser General Public License 2.1
  • Use here: number-to-word conversion used transitively by language normalization.

The installed wheel includes its LGPL 2.1 text. The Docker compliance bundle also contains the exact source distribution and a separate copy of that license.

defusedxml

  • Project: tiran/defusedxml
  • Copyright: Christian Heimes and contributors
  • License: Python Software Foundation License Version 2
  • Use here: hardened parsing for explicitly selected experimental SSML input, including rejection of DTD and entity-based XML attacks.

Model Context Protocol Python SDK

  • Project: modelcontextprotocol/python-sdk
  • Version currently used: 2.3.0
  • Copyright: the Model Context Protocol authors and contributors
  • License: MIT
  • Use here: opt-in Streamable HTTP MCP server and protocol types for local AI agent integration.

The installed distribution includes its MIT license text in package metadata.

Kokoro Vietnamese Integration

  • Project: iamdinhthuan/Kokoro-Vietnamese
  • Author: iamdinhthuan
  • License: Apache License 2.0
  • Use here: reference implementation and behavior for Vietnamese Kokoro inference.

sea-g2p

  • Project: pnnbao97/sea-g2p
  • Copyright: Pham Nguyen Ngoc Bao and contributors
  • License: Apache License 2.0
  • Use here: Vietnamese text-to-phoneme conversion used through vig2p.

vig2p

  • Package: vig2p on PyPI
  • PyPI owner: DinhThuan (iamdinthhuan on the PyPI profile)
  • Version currently used: 0.1.2
  • Use here: Kokoro-compatible Vietnamese phoneme adaptation around sea-g2p.

The vig2p package is maintained by the same identified maintainer who authored the Apache-2.0 Kokoro-Vietnamese project, and that project explicitly depends on vig2p. At the time this notice was reviewed, however, the standalone PyPI package metadata and wheel did not declare or include a license. KokoroTTS does not infer or assign a license to that separately distributed package. On 2026-10-04, the KokoroTTS maintainers requested explicit license metadata, a bundled license file, and a source-project link in Kokoro-Vietnamese issue #5. Upstream confirmation remains pending.

Model and Voice Assets

The full Docker image prefetched these assets for offline inference. The tiny image downloads an enabled model family on demand. The same licenses apply in either distribution mode.

Component Source License
Kokoro-82M model and standard voices hexgrad/Kokoro-82M Apache License 2.0
Kikiri German Martin model and voice kikiri-tts/kikiri-german-martin Apache License 2.0
Kikiri German Victoria model and voice kikiri-tts/kikiri-german-victoria Apache License 2.0
Kokoro Vietnamese model, configuration, and voices contextboxai/Kokoro-Vietnamese Apache License 2.0
Kokoro v1.1 Chinese model, configuration, and 103 voices hexgrad/Kokoro-82M-v1.1-zh Apache License 2.0

Browser Libraries

Hangry Labs Studio Components and SSML-H Tools

  • The Magic document editor, toolbar, take-bundle protocol, and dialogue script/workspace storage are adapted from faster-qwen3-tts-studio, including its Apache-2.0 components originally adapted from Hangry Labs VoxCPMTTS. Kokoro-specific changes preserve native model routing and controls.
  • Copyright: Hangry Labs and contributors. License: Apache License 2.0.
  • The engine-neutral parser is ssml-h-tools, pinned to 0.3.1, under Apache License 2.0. Its dependency uses the separately noted defusedxml terms. Distribution-supplied licenses remain in package metadata.
  • Additional Lucide glyphs use the same ISC/MIT notices below; the second locally packaged subset avoids replacing icons used by the existing UI.

WaveSurfer.js

Lucide and Feather Icons

Docker Runtime Components

The published Docker images also contain operating-system packages and language data installed during the build:

  • System eSpeak NG: a separate Debian installation in addition to the library bundled by espeakng-loader; primarily GNU General Public License v3 or later, with additional notices for specific files. The Debian package preserves its detailed copyright and license record at /usr/share/doc/espeak-ng/copyright and the GPL text under /usr/share/common-licenses/GPL-3.
  • FFmpeg: the default Debian binaries are distributed under GNU General Public License v2 or later, with individual components under compatible licenses. The installed package preserves its complete record at /usr/share/doc/ffmpeg/copyright and the corresponding license texts under /usr/share/common-licenses/.
  • Open JTalk Dictionary 1.11: includes work copyright Nara Institute of Science and Technology, the UniDic Consortium, and Nagoya Institute of Technology, distributed under three-clause BSD-style terms. The complete combined notice is retained at pyopenjtalk/open_jtalk_dic_utf_8-1.11/COPYING inside the Python runtime.

Python packages installed into the image retain the metadata and license files supplied by their distributions. Transitive packages remain governed by their respective upstream terms.

Docker License and Source Bundle

Both full and tiny images include /app/third_party with:

  • complete GPL, LGPL, MIT, Unicode-data, and Debian copyright notices for the components identified above;
  • checksum-verified corresponding-source archives for phonemizer-fork 3.3.2, num2words 0.5.14, espeakng-loader 0.2.4, and the exact eSpeak NG 1.52.0 revision used to build the loader’s bundled library;
  • loader build scripts, KokoroTTS Docker/dependency build inputs, and a generated manifest recording the installed Debian binary/source versions; and
  • SHA256SUMS plus an offline verifier.

PYTHON_PACKAGES.md records every installed Python distribution, version, declared license, and upstream URL. It intentionally reports incomplete metadata as UNKNOWN instead of guessing. NVIDIA CUDA runtime wheels used by PyTorch retain the proprietary license files supplied in their package metadata; use and redistribution of those components remain subject to their respective NVIDIA terms.

The generated manifest provides exact Debian source locations for the unmodified system eSpeak NG and FFmpeg packages. Run /app/third_party/build/verify_compliance_bundle.py inside an image, or task compliance-test after a local tiny-image build, to verify the bundle.

KokoroTTS-owned source remains under Apache License 2.0. The server runtime and Docker images are mixed-license distributions and must also be used and redistributed under the applicable third-party terms described here. The default dependency-free kokorotts HTTP client does not install these server runtime dependencies.