Licences

Nano runs entirely on the device and uses open-source AI components to do so. The notices below are required by their licences.

Built with Llama

Nano’s language generation is powered by Meta’s Llama model.

Source code access (GPL-3.0 §6): the complete corresponding source for the binaries we ship — espeak-ng (rhasspy fork), piper-phonemize, piper-wasm and phonemizer.js used by the English voice — is offered free of charge at our own address, pinned to the versions we use, for at least three years: https://seyhancanyakan.github.io/nano-biba-app/v1/source/

Turkish narration — Piper & espeak-ng

When a recorded voice-over line is not available, Nano's Turkish narration is synthesised on-device with Piper (MIT licence), using the "fahrettin" (tr_TR, medium) voice, whose training data is licensed CC0.

Piper converts text to phonemes using espeak-ng, which is licensed under the GNU General Public License v3.0 (or later). The WebAssembly build we ship is compiled via rhasspy/piper-phonemize and diffusion-studio/piper-wasm (both MIT wrappers around espeak-ng).

espeak-ng is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. Full licence text: https://www.gnu.org/licenses/gpl-3.0.txt Source code: https://github.com/espeak-ng/espeak-ng

The exact WebAssembly binary we host is unmodified from @diffusionstudio/piper-wasm@1.0.0; its build script and espeak-ng source pin live at the links above. A copy of the GPL-3.0 notice ships next to the model files: LICENSE-espeak-ng.txt.

Other components

Speech recognition, English speech synthesis and semantic matching use open-source components licensed under MIT and Apache-2.0. Vocabulary data is licensed under CC BY-SA 4.0 (CEFR-J, Octanove, YLE word lists). Details: support@sozy.io.