Privacy notice.

Silent Speech processes front-camera recordings of your face and mouth to personalize and operate a visual-speech model. These recordings are sensitive personal data. Please read this notice before participating.

1. About this notice

Silent Speech is an experimental product from Interfaces.inc, a multi-product interface lab operated by Dafdef, inc., doing business as Interfaces.inc. Dafdef, inc. is the data controller and service operator for this study. This notice describes the additional data practices specific to Silent Speech.

2. What we collect

Account and access information

We collect an account identifier from Sign in with Apple, your invitation and access status, consent records, app build and locale, and device notification tokens. Apple may provide your name or relay email address depending on the choices you make during sign-in.

Enrollment recordings

Personalized early access asks you to record 150 displayed sentences with the iPhone front camera. We process the prompted video, mouth motion, prompt text, word and capture timing, and capture-quality metadata. Two practice recordings are discarded locally.

Derived model data

We create derived mouth-region video, features, personalized model files, training status, and evaluation metrics. We do not use the recordings for face identification or to determine your identity.

Product use

When you use silent-speech chat, we process the clip needed to decode your message and the resulting text. If you explicitly submit a correction, the associated clip, prediction, correction, timing, quality metadata, and derived features may be retained under the required early-access data agreement. We do not retain server-side AI chat history in this early-access version.

Reliability information

We collect limited device, app, request, error, latency, model-version, and service-health information needed to secure, debug, and operate the study. Do not send camera recordings, access tokens, or private chat content by email when reporting a problem.

3. Why we use data

• To create, evaluate, and operate your personalized visual-speech model.
• To decode silent-speech clips you intentionally submit.
• To provide AI responses when you choose the chat feature.
• To preserve enrollment progress and notify you when training is complete.
• To protect the service, investigate failures, and measure reliability.
• To improve shared Silent Speech models and evaluate performance across participants.

Private early access is a research cohort and requires agreement to personalization, shared-model research, and contribution of product-use clips for which you explicitly submit a correction.

4. Service providers and data location

• Google Cloud hosts canonical account, recording, model, and operational data in the United States.
• Google Vertex AI processes decoded chat text when you request an AI response.
• Amazon Web Services may temporarily provide GPU compute during early access. Temporary worker copies are deleted after verified upload to Google Cloud.
• Apple provides TestFlight, Sign in with Apple, App Attest, and push-notification delivery.

We do not sell raw face or mouth recordings.

5. Retention and deletion

We retain enrollment data and personalized models while needed to provide personalization or retraining and while permitted by your recorded choices. Abandoned uploads, temporary jobs, and transient GPU copies expire sooner.

You can delete your account and data inside the app or follow the Silent Speech deletion instructions. Access is disabled immediately. Google Cloud Storage's recovery window may retain soft-deleted encrypted objects for up to seven additional days. Limited legal or security records may be retained where required.

6. Your choices and rights

You may request access, export, correction, or deletion and may stop participating at any time. Withdrawal stops new collection and future use where technically and legally feasible but cannot undo aggregate analysis or model training completed before withdrawal. A person who does not agree to every required early-access purpose cannot participate in the study.

7. Security

Data is encrypted in transit and at rest, stored in private cloud buckets, and accessed through scoped service identities. The app and API use account authentication and app attestation. No storage or transmission system is risk-free, and we cannot guarantee absolute security.

8. Eligibility and contact

This private early-access study is intended for adults aged 18 or older. For questions, rights requests, or privacy concerns, email privacy@dafdef.com.