At SXSW, AI audio startup ElevenLabs announced an ambitious 1 Million Voices Initiative, promising free voice restoration for people living with permanent voice loss. The company is inviting participants worldwide, positioning synthetic speech as a restorative tool rather than a novelty—one that can return identity, autonomy, and connection to those who’ve been silenced by illness or injury.
What The 1 Million Voices Initiative Offers
The program aims to recreate a person’s natural speaking voice using small audio samples—from voicemails, home videos, or prior recordings—and to deploy that voice in a real-time text-to-speech interface. People with ALS, head and neck cancers, post-laryngectomy patients, survivors of stroke or traumatic brain injury, and others with lasting voice impairment are the priority audience.
ElevenLabs says it is coordinating with accessibility organizations and disability nonprofits, including the Scott-Morgan Foundation, to identify eligible participants and distribute tools at scale. To spotlight early stories, the company premiered an 11-part docuseries, 11 Voices, featuring individuals who have already begun communicating with AI-rendered versions of their own voices.
A Personal Proof Point: Eric Dane’s Voice Restoration
The announcement follows a high-profile collaboration with actor Eric Dane, who worked with ElevenLabs to preserve and restore his voice while living with ALS. At SXSW, his spouse, Rebecca Gayheart Dane, described how recovering his signature sound restored confidence and offered a lasting connection for their family. Her message underscored the human stakes: a voice is not only a tool for speech—it’s memory, personality, and presence.
The docuseries profiles others experiencing similar milestones, such as a British participant with ALS who used her recreated North London accent to renew wedding vows and to reenter fast-paced conversations. These moments illustrate the difference between a generic synthetic voice and one that carries a person’s own cadence, humor, and cultural markers.
How The Technology Works For AI Voice Restoration
Modern voice restoration blends speaker cloning with expressive text-to-speech. With a few minutes of clean audio, systems can learn a speaker’s timbre and prosody; more data typically yields better nuance. Once trained, the model can render typed text as speech in near real time, enabling everything from casual chats to prepared remarks—and in some cases, live back-and-forth via a mobile device or speech-generating app.
Clinically, this extends the long-standing practice of “voice banking,” in which people at risk of voice loss record phrases ahead of time for later synthesis. Traditional banks often sounded robotic and limited in expression. Newer deep-learning methods are more natural, can preserve regional accents, and can work even when no prior banking was done—by learning from existing video or audio archives. Organizations such as the MND Association, the ALS Association, and major clinics already encourage early voice banking; the initiative could offer a lifeline to those who missed that window.
The potential impact is broad. The National Institute on Deafness and Other Communication Disorders estimates about 7.5 million people in the U.S. have a voice disorder at any given time, while the Parkinson’s Foundation reports that up to 90% of people with Parkinson’s experience voice and speech changes. Globally, the World Health Organization notes that head and neck cancers account for hundreds of thousands of new cases annually, with many patients facing long-term speech impacts after treatment.
Guardrails And Deepfake Risks In Voice Cloning
Voice cloning raises real risks, from impersonation to fraud. ElevenLabs says participation requires explicit consent, and the company has introduced detection and provenance safeguards, including an AI Speech Classifier to help identify content produced on its platform. Industry standards like the Coalition for Content Provenance and Authenticity (C2PA) are also informing approaches to labeling synthetic media.
For health-related deployments, experts emphasize strict verification, clear data retention policies, and caregiver or clinician oversight when appropriate. Limiting models to personal, noncommercial use and honoring do-not-clone requests—especially for public figures—are critical to earning public trust and ensuring accessibility benefits aren’t overshadowed by misuse.
Why Scale Matters For Assistive Voice Technology
Reaching 1 million people would mark a step change in assistive communication. Traditional speech-generating devices can cost thousands of dollars and often rely on generic voices. Software-based systems that reproduce an individual’s own sound can be distributed more widely, updated over time, and localized across languages and dialects—key for communities underserved by existing assistive tech.
Success will hinge on logistics: onboarding in clinical and community settings, multilingual support, safeguards for minors and vulnerable users, and training for speech-language pathologists who can integrate AI voices into therapy. If executed well, the initiative could normalize personal voice restoration in the same way screen readers normalized digital accessibility—quietly, pervasively, and for the better.
For now, ElevenLabs is soliciting candidates and partners to build a global pipeline. The promise is simple but profound: give people back the sound of themselves, at scale, and let technology do what it does best—amplify human dignity.
