Voice sources and licenses

Optional voices generate speech on your device. The optional Kitten package downloads from the same host as Ride. No direction text is sent to a speech service. Device voices are provided by your operating system; in Local only mode Ride permits only voices the browser identifies as local.

Downloads are verified against pinned SHA-256 digests. Generated audio stays in bounded memory and is discarded when the route or voice changes. Downloads can be deleted in the Voice settings.

KittenTTS Nano 0.8 int8

Model and voices, Apache-2.0. The tokenization and text preprocessing are adapted from KittenTTS Web 0.1.0, Apache-2.0. Ride uses a modified direct-inference adapter; no analytics SDK is included. License.

English pronunciation

The CE phonemizer is KittenML’s original Apache-2.0 implementation. Its compiled runtime is extracted from the pinned KittenTTS Web bundle without behavior changes. It is not the unidentified eSpeak executable bundled in some other JavaScript TTS packages.

English rules and dictionary by Jonathan Duddington, Reece H. Dunn and eSpeak NG contributors are distributed in unmodified source form in the optional package under GPL-3.0-or-later. License.

ONNX Runtime Web

Version 1.26.0, Microsoft Corporation. MIT license and third-party notices. CPU/WASM execution uses one inference thread. No GPU or microphone permission is requested.

Ride voice adapter source · Pinned asset manifest

Neural voices and pronunciations can make mistakes. Browser audio and GPS may pause when the screen locks. Planned-route guidance uses the downloaded routing graph; road data can be incomplete or outdated.

Back to Ride