Public disclosure
Tadamo uses artificial intelligence.
Last updated 2026-07-13. Applies to the Tadamo private family preview; it will be revised and re-noticed before any family outside the developer's own uses the app.
What Tadamo is
Tadamo is an assistive communication app. It uses speech recognition and a child-specific phrase and pronunciation context to suggest intended text, then shows and can speak that text for a listener. It is not a medical device, diagnosis, treatment, emergency service, or substitute for a qualified professional.
Which systems run
| System | Purpose | Information used |
|---|---|---|
| Microsoft Azure Speech | Near-real-time speech recognition and, only after enough verified samples, a child-specific speech model. | Audio recorded after the child starts Talk, plus text phrase hints and measured pronunciation patterns. The device receives a short-lived Azure token; Tadamo never sends the Azure resource key to the device. |
| On-device speech synthesis (Apple) | Speaks text after the speaker taps Speak. Automatic output would additionally require guardian voice permission and a current validation policy bound to the exact speaker and active personal model; Tadamo does not issue that automatic-output policy in this release. | Runs on the device. The output text does not need to leave the phone for this feature. |
When speech processing runs
- Only after the child starts Talk; Tadamo has no wake word and is not always listening.
- A visible recording state is shown while the microphone is active.
- Azure processing is denied unless the guardian signed the current agreement and granted both speech processing and third-party disclosure for that child.
- Revocation blocks new Azure tokens. The offline phrase board remains available.
Speech output
- Every live interpretation is caption-only by default.
- With guardian voice permission, the speaker may review the caption and tap Speak.
- Automatic speech requires both that permission and a separately issued, current validation policy bound to this speaker and the active personal model. This release issues no such policy, so it does not automatically speak live interpretations.
- Turning voice permission off suppresses both manual and automatic synthesized translation output.
Learning safeguards
- A caption or spoken interpretation is logged as unverified and cannot teach the personal model.
- A guardian can correct text-only live history in Today's Talk; only that explicit correction may teach a private heard-to-intended word pattern.
- Saved game or calibration recordings start quarantined and cannot train until a guardian confirms or corrects the label.
- Candidate personal models must improve on a separate frozen evaluation set before promotion.
- AI can be wrong. The child or guardian can retry, correct, exclude, or delete a recording.
Agreement and contact
Read the guardian agreement, children's privacy notice, and retention schedule. Questions: support@playstarseeker.com or 727-638-3936.