Workshop · Note 7
Baba's voice: synthetic voices and AI translation.
Why Katun says openly that the voices are synthetic and the translations come from an AI.
The Katun page says it openly: the translations were made with AI, and Baba's voice and the tellers' voices are synthetic. Here is why, how we check, and what it means for you as a listener.
Why several languages
Katun comes in English, Français, Italiano, Español, Deutsch, Srpski (latinica and ћирилица) and Hrvatski. You choose the language on the first launch and can change it any time. The tales exist in each of these languages, and Baba reads them aloud in English, French, Italian, Spanish and German. In Serbian and Croatian there is only the text: we found no voice model for them that runs on our own computer, sounds good enough and has a licence that allows its use in an app.
Why synthetic voices
Baba should sound warm, deep and unhurried, with a bit of a smile in her voice. We decided on generated voices. One reason is that they make more languages affordable: the real work then lies in the translation, not in the recording.
The voices are designed, not copied: we describe a voice, and the voice model generates it. Rebuilding a real person would not have come into question without their written consent. The model runs on our Mac. The recordings are made in advance and ship as audio files in the app. Nothing is generated on your device, and no data from players goes to the voice tool. Each language has its own reference recording, so that Baba does not speak German with an English accent.
How we check
The translations follow a style guide with a glossary. Names and katun words such as Baba, kajmak or sijelo stay the same in every language, and we neither shorten the tales nor add to them. After that, a second, independent review reads each language against it: is Baba's tone right, are the placeholders intact, is anything missing.
For the voice, an automatic speech check listens to every single recording. If what it hears differs from the text, the line goes on a list to be listened to again: does a word sound wrong, mispronounced or unnatural? Often the check is wrong, sometimes the voice is.
What it means for you
If you do not play Katun in English, the app tells you that the translation was made with AI and asks you: if something reads oddly, write to us at support@komovi.ch. The voices are synthetic and usually sound calm and clear, but not like a human being. We would rather say so beforehand. The tales themselves are freely retold from the region's folk tales. If something strikes you in any language, every line helps us.
More notes
KatunWhy Katun existsA summer pasture, a herd, a fire in the evening: how the idea for Katun came about, and why nothing nags.Read - KatunDrawing a mountain in codeKatun's pictures come from the code of the scene: paper by day, hearthlight at night.Read
KomoviNo server: what it costs us and what you getNo account, no server of our own: what that means for backup, support and statistics, and what you get.Read