Asset credits and licences

StoryToon’s 100 characters and 26 places are made in-house from code, so we own them fully. These are the third-party pieces we ship, each with a licence that allows commercial use.

Last updated 21 October 2026

AssetSourceLicenceNotes
onnxruntime-web 1.30.0 (npm, voice)https://github.com/microsoft/onnxruntimeMITWASM runtime that runs Piper voices in the browser. Bundled deps: onnxruntime-common (MIT), flatbuffers (Apache-2.0), long (Apache-2.0), guid-typescript (ISC), platform (MIT), protobufjs + @protobufjs/* (BSD-3-Clause).
eSpeak-NG WASM build: @echogarden/espeak-ng-emscripten 0.3.5 (espeak-ng.js + espeak-ng.data)https://www.npmjs.com/package/@echogarden/espeak-ng-emscripten ; source https://github.com/echogarden-project/espeak-ng (branch fork), build https://github.com/echogarden-project/espeak-ng-emscripten, upstream https://github.com/espeak-ng/espeak-ngGPL-3.0Robotic voices, 100+ languages (ur, hi, ar ...) and the phonemizer for Piper. Shipped as separate, unmodified files (24 MB data + 0.8 MB JS, wasm2js build), loaded at runtime from a URL, never bundled or modified; COPYING + SOURCE.txt are served next to them (packages/voice/scripts/fetch-models.mjs).
Piper voice en_US-ljspeech-medium (63.5 MB)https://huggingface.co/rhasspy/piper-voices (rev c10ece1a), model by Bryce Beattie https://brycebeattie.com/files/tts/Model: public domain (author) / repo MIT; dataset LJ Speech: public domainTrained from scratch on LJ Speech. US female.
Piper voice en_US-kristin-medium (63.5 MB)same repo; https://brycebeattie.com/files/tts/Model: public domain; dataset: LibriVox recordings (public domain)Trained from scratch. US female.
Piper voice en_GB-cori-medium (63.5 MB)same repo; https://brycebeattie.com/files/tts/Model: public domain; dataset: LibriVox (public domain)Trained from scratch. UK female.
Piper voice en_US-norman-medium (63.5 MB)same repo; https://brycebeattie.com/files/tts/Model: public domain; dataset: LibriVox (public domain)Trained from scratch. US male.
Piper voice en_US-john-medium (63.5 MB)same repo; https://brycebeattie.com/files/tts/Model: public domain; dataset: LibriVox (public domain)Fine-tuned from kristin (public domain). US male.
Piper voice en_US-bryce-medium (63.5 MB)same repo; https://brycebeattie.com/files/tts/Model: public domain; dataset: the author's own recordings, released public domainFine-tuned from the author's own unreleased voice. US male.
Kokoro-82M v1.0, 8-bit ONNX onnx/model_quantized.onnx (92.4 MB, sha256 fbae9257…1478) + voice style files voices/<id>.bin (0.5 MB each: af_heart, af_bella, af_nicole, af_sarah, am_michael, am_fenrir, am_puck, bf_emma, bm_george, bm_fable, hf_alpha, hf_beta, hm_omega, hm_psi, ef_dora, em_alex, ff_siwis, if_sara, im_nicola, pf_dora, pm_alex)https://huggingface.co/onnx-community/Kokoro-82M-v1.0-ONNX (rev 1939ad2a), original weights https://huggingface.co/hexgrad/Kokoro-82MApache-2.0 (weights and voices, both repos; checked 2026-11-13)Natural free voices since 2026-11-13 (packages/voice/src/kokoro.ts): English (US/UK), Hindi, Urdu (via the Hindi voices), Spanish, French, Italian, Portuguese. Model card: trained only on permissive / non-copyrighted audio (public domain, Apache/MIT-licensed audio, CC BY 4.0 SIWIS French (<11 h; attribution: SIWIS database, Idiap) and CC BY 3.0 Koniwa Japanese), plus synthetic audio from closed commercial TTS providers. Residual note: that synthetic share is governed by those providers' terms toward the model's author, not toward users of the Apache-2.0 weights. Phonemes from our eSpeak-NG module (no extra G2P code shipped). Served from R2 voices/kokoro/*, cached offline.
Piper voices for other languages (re-checked 2026-11-13)https://huggingface.co/rhasspy/piper-voices-Kokoro now covers es, fr, it, pt, hi. No Urdu, Arabic or Hindi Piper voice has commercial-use data. de_DE-thorsten (CC0 data) stays a candidate: its training lineage (scratch vs fine-tuned from lessac) is not documented clearly enough to ship.
three.js 0.186.1 (npm three, 3D renderer)https://github.com/mrdoob/three.jsMITBundled into the studio by @storytoon/engine3d (GLTFLoader, DRACOLoader from examples/jsm).
Draco decoder (three.js examples/jsm/libs/draco: draco_decoder.wasm, draco_wasm_wrapper.js)https://github.com/google/draco (via https://github.com/mrdoob/three.js)Apache-2.0Shipped unmodified in packages/engine3d/draco/, served at /storytoon/files/3d/draco/; decodes the Draco-compressed glb files.
Noto Nastaliq Urdu (Regular), Noto Naskh Arabic (Regular, Bold), Noto Sans Devanagari (Regular, Bold)https://github.com/notofonts/notofonts.github.io (unhinted TTF; projects notofonts/nastaliq, notofonts/arabic, notofonts/devanagari)SIL OFL 1.1Public site only (apps/site/fonts/src/ + OFL.txt). The site build subsets them to the characters of the ur/ar/hi pages (woff2, subset-font, BSD-3-Clause, build-time only) and loads each only on its language's pages. Font names are kept (OFL: subsetting is allowed; the files are not sold on their own).
Noto Sans Bengali (Regular, Bold; static instances wght 400/700 of the v2.001 variable font)https://github.com/notofonts/bengali (Google Fonts NotoSansBengali-VariableFont_wdth,wght; local copy taken from the font folder of the installed Blender 4.5 LTS, whose licence list names Noto as OFL 1.1)SIL OFL 1.1Public site only (apps/site/fonts/src/ + OFL.txt), used by the bn pages exactly like the fonts above. Instanced to two static weights with subset-font (HarfBuzz), every character and the name/licence table kept; Noto has no Reserved Font Name, so the family name stays.
Fredoka (variable font Fredoka[wdth,wght].ttf, v2.001; used as one static SemiBold instance, wght 600 / wdth 100)https://github.com/google/fonts/tree/main/ofl/fredoka (Google Fonts; upstream https://github.com/hafontia/Fredoka-One, designers Milena Brandão, Hafontia), downloaded 14 Nov 2026SIL OFL 1.1 (no Reserved Font Name)The v3 look's rounded display font (headings, buttons, brand). Public site: apps/site/fonts/src/Fredoka-VF.ttf + Fredoka-OFL.txt; the build subsets it to Latin + Latin Extended-A and common punctuation (16 KB woff2, subset-font, name/licence table kept); it is loaded only on v3 pages, after the page has loaded, so v1 / v2 pages never download it. Studio home: the same subset committed as apps/web/src/fonts/fredoka-600-latin.woff2 + OFL.txt, loaded only by the v3 dashboard rules (and the log-in / sign-up window, apps/web/src/account/authWindow.css). Account pages (sign-up, forgot, reset, verify): the same bytes embedded in workers/api/src/account/fredoka.ts and served by the Worker at /storytoon/account/fonts/fredoka-600.<hash>.woff2 (CSP font-src 'self'; a unit test checks the bytes match the committed file).
lamejs 1.2.1 (npm lamejs, lame.min.js only: MP3 encoder for the converters)https://www.npmjs.com/package/lamejs ; source https://github.com/zhuker/lamejs ; LAME https://lame.sourceforge.io/LGPL-3.0 per npm (LAME: GNU Library/Lesser GPL; the package's COPYING text is served)Shipped like eSpeak-NG: a separate, unmodified file at /storytoon/app/vendor/lamejs/lame.min.js, loaded at run time only when someone converts to MP3, never bundled or modified; LICENSE, COPYING and SOURCE.txt are served next to it (packages/media/vendor/vite-plugin.mjs serves them in dev/preview and emits them unchanged into the studio build). devDependency of @storytoon/media; its npm dependency use-strict (ISC) is not shipped.
mp4-muxer 5.2.2 (npm)https://github.com/Vanilagy/mp4-muxerMITBundled code: writes MP4 / M4A files in the browser (@storytoon/export video export, @storytoon/media editor export and M4A converter).
gifenc 1.0.3 (npm)https://github.com/mattdesl/gifencMITBundled code: animated GIF encoder (@storytoon/export GIF export, @storytoon/media video-to-GIF converter).

eSpeak NG is GPL-3.0 software. It is shipped as separate, unmodified files with its licence text and a link to its source code, and loaded at run time. Music and sound effects are made in code by StoryToon.