Made with
a little care.

Original soundscapes

Every ambient layer, chime, tone, and mix is generated with original procedural synthesis in your browser. No commercial music or sampled nature recordings are used.

Morning voices

The original SOMNIA scripts were rendered at build time using Kokoro-82M v1.0 by hexgrad, an Apache-2.0 licensed text-to-speech model. Voice packs: af_heart (Sweetheart), af_bella (Motivator), bf_emma (Comedian), af_sky (Savage). The model and its code are not distributed in the app. No inference service, voice cloning or paid AI API is used.

Kokoro credits its StyleTTS 2 and ISTFTNet foundations. Its model card acknowledges Koniwa (CC BY 3.0) and SIWIS (CC BY 4.0) among its training sources. See the model’s attribution and acknowledgements.

Your own recordings belong on your own device. SOMNIA does not upload them.