A moss-covered stone lantern standing in a bamboo grove

Guide production

People and AI finish an audio guide in one flow.

From the chapter outline to the manuscript, proofreading, audio and misreading checks. Work is split between what AI does and what people decide, moving toward a guide that is ready to publish.

Talk to us about WOUDiOFeatures

The production flow

Each step is done together — not by AI alone, and not by people alone.

  1. Research first

    Facts are recorded with source URLs, aiming for 8 to 10 per chapter. Numbers and years are cross-checked against a second source.

  2. A chapter outline

    People and AI decide the chapters and what each should convey. The AI proposes; people decide.

  3. Choose a storyteller

    Pick a storyteller (a voice and style preset) that fits the subject — an analytical one for structures and techniques, a historian for people and history — per venue.

  4. Write the manuscript

    AI writes a draft and people edit it. Every version is kept and can be restored. Translation, and proofreading of the translation, happen in the same flow.

  5. AI proofreads

    AI reads the manuscript as an editor and scores it; anything below the bar is held back. It checks facts against the venue’s own materials and lists numbers and years that the research notes do not contain.

  6. Apply the fixes

    Proofreading findings can be applied to the manuscript with the version kept. You choose whether fixes are applied automatically or only proposed. The fixed text is checked again.

  7. Generate the audio

    Audio is generated with a speech engine that fits the storyteller. You can see a cost estimate before regenerating.

  8. AI listens and checks

    AI listens to the generated audio, compares it with the manuscript, and lists misreadings, each with a severity.

  9. Record in the dictionary, regenerate

    Serious misreadings are recorded in the venue’s pronunciation dictionary, and only that guide and language are regenerated and checked again (with a cap on repeats). Anything that needs a person goes to a review queue.

  10. Decide whether it can be approved

    Only guides that meet every bar — the editorial score, the audio check, and audio that matches the latest manuscript — are approved automatically; the rest go to a person. If the manuscript is edited later, the older audio is no longer eligible for automatic approval.

What keeps readings accurate

Who does what

AI drafts, proofreads, listens to the audio and lists misreadings. People decide the outline, choose the storyteller, edit the manuscript and judge what needs a closer look. An AI check is not a substitute for a person’s review.

Frequently asked questions

Is an AI-written manuscript published as is?
No. Only guides that meet every quality bar (the editorial score, the audio check, and audio that matches the latest manuscript) are approved automatically. The rest go to a person.
Will misreadings disappear?
We cannot promise that. AI listens, lists misreadings, records them in the dictionary and regenerates, which reduces them. If AI misses one, a person can review and fix it.
How are names and places corrected?
Add them to the pronunciation dictionary, or specify the reading inline with ruby. Serious misreadings found by the audio check are recorded in the venue’s dictionary automatically; promoting an entry to the shared dictionary is a person’s decision.
How are multiple languages produced?
From the source manuscript, with translation and proofreading of the translation. Each language gets its own manuscript and audio checks.

More features

Traditional townhouses with paper lanterns along the eaves

Talk to us about adopting WOUDiO

A rough idea of your venue and what you want to do is enough. We will get back to you.

Sending opens your email app with the message ready.