Preserving vocal identity is an essential accessibility capability for individuals facing degenerative speech conditions, ALS, or vocal cord trauma. In iOS 18, Apple leverages on-device machine learning and the Neural Engine to synthesize a personalized vocal facsimile—called Personal Voice—from just 15 minutes of recorded speech prompts. Personal Voice integrates directly with Live Speech across FaceTime, standard phone calls, and in-person conversations while safeguarding biometric voice models behind on-device cryptographic encryption.
Privacy Architecture and On-Device Processing
Apple engineered Personal Voice with strict cryptographic safeguards:
- On-Device Machine Learning: The voice synthesis neural model trains entirely on the device processor while the iPhone is connected to power and locked overnight. Raw audio recordings never leave the device or upload to iCloud without explicit end-to-end encrypted backup consent.
- Face ID Authentication: Generating, editing, or exporting your voice model requires Face ID or Touch ID authentication.
- Live Speech API Integration: Third-party communication apps can synthesize your voice through standardized Apple accessibility frameworks without accessing raw training data.
Recording the Personal Voice Training Dataset
To train your personal voice model on an iPhone running iOS 18:
- Find a completely quiet indoor room free of background air conditioning, echoes, or television noise.
- Open the Settings app on your iPhone.
- Navigate to Accessibility and scroll down to the Speech section.
- Tap Personal Voice, then tap Create a Personal Voice.
- Authenticate with Face ID or your device passcode.
- Hold your iPhone approximately 6 to 12 inches (15–30 cm) from your face at natural speaking volume.
- Read through the 150 short phrases presented on the display screen. The system automatically advances upon detecting complete, natural phrase enunciation.
- If you make a mistake or cough, tap the rewind arrow to re-read the active prompt.
Synthesizing the Voice Model Overnight
Once audio capture finishes, the Neural Engine processes the acoustic parameters:
- Connect your iPhone to its official MagSafe or USB-C charger.
- Lock the display and ensure the device remains connected to Wi-Fi.
- The training process runs during idle background states to prevent device throttling, typically completing within 4 to 8 hours depending on processor generation (A16, A17 Pro, or A18).
- A system notification alerts you once synthesis is complete: Your Personal Voice is ready to use.
Using Live Speech in Phone and FaceTime Calls
To deploy your voice model during active phone conversations or real-world discussions:
- Go to Settings > Accessibility > Live Speech and toggle the feature on.
- Under Voices, select your newly synthesized Personal Voice.
- Triple-click the physical Side button (or Power button) anywhere in iOS to open the Live Speech overlay.
- Type your message or select from customized quick phrases. Tap Send or the audio icon to speak the phrase out loud through your iPhone speakers or directly into active FaceTime and phone calls.