AI Accessibility

Voice Cloning for People with Speech Impairments

By EZUD Published · Updated

Voice Cloning for People with Speech Impairments

Losing your voice changes how you are perceived. People with progressive conditions like ALS (amyotrophic lateral sclerosis), Parkinson’s disease, and certain cancers that affect the vocal tract face the prospect of communicating through generic synthetic voices that sound nothing like them. AI voice cloning changes that equation, allowing people to preserve their vocal identity or create a personalized synthetic voice that maintains their sense of self.

How Voice Cloning Works

Voice cloning uses machine learning to create a digital model of someone’s voice. The user provides voice samples (reading passages aloud), and the AI extracts vocal characteristics: pitch, tone, cadence, pronunciation patterns, and acoustic texture. The resulting model can then generate new speech in that voice from any text input.

The amount of voice data needed has dropped dramatically. Early systems required hours of studio recordings. Current technology can create a usable voice clone from as little as 15 minutes of reading.

Apple Personal Voice

Apple introduced Personal Voice in iOS 17 as an accessibility feature specifically designed for people at risk of speech loss. Users read a series of text prompts aloud to their iPhone or iPad for approximately 15 minutes. On-device machine learning processes this audio to build a vocal model.

Once created, Personal Voice integrates with Live Speech, allowing users to type text that is spoken aloud in their own voice during phone calls, FaceTime conversations, and in-person interactions. All processing happens on-device, keeping voice data private.

Personal Voice works on iPhone, iPad, and Mac. For users who are already unable to speak, Apple also offers pre-built voices, though these are generic rather than personalized.

Other Voice Preservation Options

VocaliD (now Acapela Group)

VocaliD pioneered custom synthetic voices by blending a user’s residual voice (even if impaired) with a donor voice of similar characteristics. The result is a unique voice that preserves individual vocal traits even when the user’s speech is no longer fully clear.

ModelTalker

Developed at the Nemours Alfred I. duPont Hospital for Children, ModelTalker is a free voice banking system. Users record a set of phrases through a web interface, and the system generates a personalized synthetic voice for use with AAC devices.

SpeakUnique

SpeakUnique, developed at the University of Edinburgh, creates synthetic voices from as little as a few minutes of speech. It is specifically designed for people with dysarthria (slurred speech) and can work with impaired voice samples.

Resemble AI

Resemble AI provides an API for custom voice creation. While marketed primarily for media and entertainment, its technology has accessibility applications for creating personalized voices from limited recordings.

When to Start Voice Banking

The medical consensus is clear: start voice banking as early as possible after diagnosis of a progressive condition. Voice quality degrades as conditions advance, and the best voice clones come from clear, full-range recordings. Speech-language pathologists increasingly include voice banking in early-stage treatment plans for conditions like ALS.

For people who have already lost significant vocal function, tools like SpeakUnique and VocaliD can work with impaired samples, though the resulting voice will be less natural than one created from healthy speech.

Privacy and Ethical Considerations

Voice cloning raises obvious concerns about impersonation and fraud. Accessibility use cases require safeguards:

  • On-device processing (as Apple implements) prevents voice data from leaving the user’s device.
  • Consent frameworks ensure the voice is created and used only with the individual’s explicit permission.
  • Watermarking embeds inaudible markers in synthetic speech to identify it as AI-generated, a technical countermeasure against voice fraud.
  • Access controls prevent unauthorized use of someone’s voice model.

For related reading on speech-to-text technology, see speech-to-text accuracy comparison 2026. For the broader landscape of AI in accessibility, see the AI accessibility guide.

Key Takeaways

  • AI voice cloning allows people facing speech loss to preserve their vocal identity rather than communicating through generic synthetic voices.
  • Apple Personal Voice requires only 15 minutes of recording, processes entirely on-device, and integrates with Live Speech for calls and conversations.
  • Specialized tools (VocaliD, ModelTalker, SpeakUnique) serve users at different stages of speech loss, including those with already-impaired voice quality.
  • Voice banking should begin as early as possible after diagnosis of progressive conditions for the best results.
  • Privacy protections including on-device processing, consent frameworks, and watermarking are essential to prevent voice fraud.

Sources

More in AI Accessibility

AI Is Transforming Accessibility: The Complete Guide Guide AI Navigation Assistance for Visually Impaired Users AI Accessibility Research Frontiers Speech-to-Text Accuracy Comparison (2026)

View all AI Accessibility articles →