Amyotrophic lateral sclerosis (ALS), better known as Charcot's disease, affects approximately 2 to 3 people per 100,000 worldwide and leads to a progressive loss of motor function, including speech1. For many patients, this loss of voice is one of the most difficult changes to cope with, as it directly affects communication, independence, and social connections. In response to this challenge, artificial intelligence is opening up new possibilities. At the start of this year, the French lab Kyutai introduced Invincible Voice, a voice AI developed in collaboration with Olivier Goy—a committed entrepreneur who himself has ALS—to enable patients to continue communicating despite the loss of speech.
When AI Becomes a Tool for Connection and Independence
The loss of the ability to speak affects the vast majority of people with ALS as the disease progresses. Clinical studies estimate that more than 80% of patients develop severe communication difficulties in the advanced stages2. Invincible Voice is part of an approach in which AI is not viewed as a mere technical tool, but as a means of preserving relationships with others. By involving researchers and patients from the very beginning of the design process, Kyutai places real-world use and human impact at the heart of technological development.
Recreate a voice using just a few seconds of audio footage
From a scientific standpoint, “Invincible Voice” is based on major advances in neural speech synthesis. Kyutai notes that it is possible to reconstruct a voice from approximately ten seconds of recording, a threshold made possible by recent advances in deep learning models3. According to Olivier Goy, the teams relied on audio archives recorded before the onset of the disease, particularly podcasts. This method allows for the reproduction not only of words, but also of timbre, intonations, and rhythm—essential elements of vocal identity.
How "Invincible Voice" Works, Step by Step
“Invincible Voice” is based on Unmute, an open-source architecture developed by Kyutai. The system operates in a modular fashion. It begins with a real-time transcription of the patient’s conversation partner’s speech. A language model then analyzes the context and can suggest several appropriate responses. Finally, the selected response is synthesized vocally using the patient’s reconstructed voice, resulting in a smooth and natural delivery. This comprehensive workflow enables continuous voice interactions without any artificial breaks in the conversation.
Accessibility designed for people with severe disabilities
The tool was designed to accommodate various levels of motor impairment. For patients with severely limited mobility, “Invincible Voice” is compatible witheye-tracking devices, allowing users to select responses with their gaze. These assistive technologies are now recognized as essential tools for maintaining communication, particularly in neurodegenerative diseases4. By integrating these devices from the design stage, Kyutai significantly expands the pool of potential beneficiaries.
A strategic choice: open source in the healthcare sector
Kyutai has chosen to release “Invincible Voice” as open source, to enable the scientific and medical community to adopt the technology. This openness facilitates the adaptation of the tool to other speech disorders, such as post-stroke aphasia or certain neurodegenerative diseases. The modular architecture allows each component—transcription, text generation, or speech synthesis—to be improved independently, fostering distributed and collaborative innovation.
A French voice AI ecosystem is taking shape
The project is supported by Gradium, Kyutai’s first spin-off specializing in speech synthesis. Officially launched recently, the company has raised €60 million to commercialize its technologies and build a French ecosystem for voice AI5. This momentum bridges the gap between fundamental research, industrial innovation, and concrete medical applications—a key challenge for the large-scale deployment of assistive solutions.
Ethical issues, consent, and responsible use
The ability to mimic a human voice raises significant ethical questions. As part of “Invincible Voice,” Kyutai emphasizes strictly authorized and medical use, where speech synthesis aims to restore a lost ability, not to impersonate someone. The project’s open-source nature also contributes to transparency and oversight by the scientific community. Research on AI ethics indeed emphasizes that voice synthesis technologies must be regulated to prevent misuse, while maximizing their positive social impact6.
To give a voice back, both literally and figuratively
With “Invincible Voice,” Kyutai and Olivier Goy demonstrate that artificial intelligence can play a decisive role in the medicine of tomorrow. By giving a voice back to people gradually silenced by illness, this project illustrates a human-centered vision of AI, where technological prowess serves dignity, inclusion, and social connection. More than just a tool, Invincible Voice embodies a different approach to innovation, guided by its real-world impact on patients’ lives.
Learn more
The use of artificial intelligence to restore or enhance human capabilities is part of a broader trend of applying AI to healthcare. On a related topic, check out our article “MedGPT: The Free French Medical AI That Rivals ChatGPT”, which analyzes how specialized models are emerging to support patients and healthcare professionals, while raising key questions about reliability, ethics, and sovereignty.
References
1. World Health Organization. (2023). Motor neuron diseases.
https://www.who.int
2. Hardiman, O. et al. (2017). Amyotrophic lateral sclerosis. The Lancet. https://doi.org/10.1016/S0140-6736(17)31287-4
3. Taylor, J. et al. (2022). Neural Voice Cloning with Limited Data. arXiv.
https://arxiv.org/abs/2203.00248
4. Lancioni, G. et al. (2020). Assistive technologies for people with severe motor disabilities. Research in Developmental Disabilities.
https://doi.org/10.1016/j.ridd.2020.103722
5. Les Échos. (2024). Gradium raises €60 million to develop voice AI.
https://www.lesechos.fr
6. Floridi, L. et al. (2018). AI4People—An Ethical Framework for a Good AI Society. Minds and Machines.
https://doi.org/10.1007/s11023-018-9482-5 /a>