How we make it with AI

Sovalu is a small family studio. Texts, translations and voices are created by artificial intelligence and checked by a separate AI reviewer. We never hide the AI.

Who’s behind it

We’re Jan and Aneta, and we have two kids, who were five and seven when we started. We want to make calm stories with no chases and no monsters, the kind that are easy to drift off to.

Jan is an IT engineer. He builds the tech, from the tools we use to create the stories with the help of AI to this website, and he looks after episode production. Aneta is the voice of Sovalu. Episodes read by an AI clone of her voice are still in preparation; no published episode uses her voice yet. The clone of her voice was made with her consent and we use it only for Sovalu.

We write the texts with the help of AI and we’re open about it. And if Aneta ever says stop, we’ll stop using her voice.

From idea to listening

  1. Idea and text. AI writes the text following fixed rules: a slow pace, nothing scary for children, and a wind-down at the end.
  2. Translation. AI prepares the translation and a second AI checks it the way a native speaker would. If you spot a mistake, let us know.
  3. Voice. Stories are read by a synthetic voice. If we use a voice based on a real person, it is only with their consent, and we always say so.
  4. Review. Before publishing, every text is checked by a separate AI reviewer against a fixed checklist. No person reads the texts before they are published. If you spot a mistake, let us know and we’ll fix it.
  5. Publishing. We are setting it up so that one reviewed record becomes a page on this website, an item in our podcast RSS feed and a video with a calm picture.

Whose voice you hear

We want the voice reading to you in the evening to belong to someone who really stands behind the project and vouches for what you hear. That is why we are preparing episodes read by an AI clone of Aneta’s voice. Every episode that uses it will say so on its card and in the player: “Read by Aneta · AI voice”.

No published episode uses Aneta’s voice yet. The test recording of The Path to the Mountain Lake uses Kami, a voice from the ElevenLabs voice library.

How we label AI

  • Labels on every card and next to the text: how the text, voice and translation were made.
  • One sentence at the start of each episode saying the voice and the text were created by AI.
  • The first line of every description in podcast apps and on YouTube.

Where to find us

Every episode’s home will be this website. Podcast apps (Spotify, Apple Podcasts, Amazon Music, Pocket Casts and others) will pick up new episodes from our RSS feed by themselves. The feeds are still in preparation.

On YouTube we are preparing a channel for adults and a separate kids’ channel. Videos will be marked as synthetic content, and children’s stories also as made for kids.

What we never do

  • We never promise treatment or “guaranteed sleep”.
  • We never use other people’s characters or brands, only our own world.
  • We only use a voice based on a real person with that person’s consent.
  • We never pass off an AI voice as a live reading.
  • We base our articles about sleep on citations of peer-reviewed studies.