Third-party notices

Effective 11 July 2026

Tavolo's AI voice host runs on open-source speech models that we self-host. Several are published under licences that require attribution. This page is that attribution, and a plain statement of what speaks when you call a restaurant using Tavolo.

Speech synthesis — the voice you hear

German and Dutch voices: Piper voice models trained on Multilingual LibriSpeech (MLS), by Pratap et al., available at openslr.org/94 and licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). We use these models unmodified.

English voice: a Piper voice model trained on the LJ Speech Dataset by Keith Ito, which is in the public domain.

English fallback voice: Kokoro-82M by hexgrad, licensed under Apache License 2.0.

Speech synthesis engine: Piper, licensed under the GNU General Public License v3.0. Tavolo operates it as a self-hosted network service and does not distribute it.

Speech recognition — understanding what you say

Whisper by OpenAI and the faster-whisper implementation by SYSTRAN, both licensed under the MIT Licence.

Why these particular models

Tavolo only uses speech models whose licences permit commercial use, and checks how each model's weights were produced rather than relying on the licence label alone. A model can carry a permissive licence and still be trained from a base whose data is restricted to research; where that is the case, we do not use it.

This is a good-faith summary for Tavolo's pilot, not legal advice. Questions or data requests: hello@tavolo.io.

Back to home