How Your Voice Is Handled
Effective date: 14 September 2026
Your microphone audio stays on your iPhone
Unsaid needs to hear you so it can turn your words into text, but it does not upload your microphone audio. Apple’s on-device speech recognition performs transcription on your iPhone while you hold to speak.
If supported on-device recognition is unavailable for the selected language, Unsaid does not silently switch to cloud transcription. The practice cannot start until on-device recognition is available.
What happens during a practice
1. You hold to speak. The microphone provides short-lived audio buffers to Apple’s on-device speech recognizer.
2. Your iPhone creates text. The temporary microphone audio is not written to your practice history, iCloud, app logs, analytics, or the Unsaid backend.
3. Unsaid sends text, not your voice. The transcript text and the minimum scenario context needed for the current turn are sent securely to Unsaid’s backend and OpenAI for moderation and an in-character reply.
4. The partner’s reply becomes audio. ElevenLabs receives the AI-generated reply text and creates the practice partner’s voice. ElevenLabs does not receive your microphone audio.
5. You hear the reply. Generated audio is returned for playback in memory. Unsaid does not save that audio in local history, iCloud, or its backend database.
6. You choose what to keep. When the practice ends, Unsaid can save the transcript, summary, and feedback on your device. If you enable private iCloud backup, supported history can sync to your private iCloud database.
What leaves your device
The text of your conversation can leave your device because online AI services need it to respond and create feedback. The current scenario, goals, and recent text turns may also be included. Unsaid limits requests to the content needed for the operation and designs production logs not to contain ordinary transcript text.
Do not include names, account numbers, passwords, addresses, medical records, or other private details unless they are genuinely necessary. Usually, a role such as “my manager” or “my partner” is enough.
Provider retention
Unsaid does not store ordinary transcript text or generated audio in its backend database. OpenAI requests are configured with storage disabled for the application workflow, but OpenAI’s default API abuse-monitoring logs may retain request content for up to 30 days unless different approved controls apply. ElevenLabs may process or temporarily retain generated reply text and generated audio under the retention settings and API terms that apply to Unsaid’s account.
Provider retention is not the same as Unsaid saving a practice to its own database. We use no-retention or deletion controls where they are available and verified, and we will update this page if those settings materially change.
Your controls
• Revoke microphone or speech-recognition access in iOS Settings.
• Delete one practice or all practice history in Unsaid.
• Turn private iCloud backup on or off in Unsaid Settings.
• Delete your Unsaid account from Settings.
• Email thechillguy600@gmail.com with a privacy question.
The honest summary is simple: your microphone audio stays on your iPhone; the transcript text is processed online so the AI partner can answer.