Researcher of Algerian origin wants voice to replace keyboards

Neil Zeghidour, a researcher of Algerian origin, leads Gradium, a Paris artificial intelligence start-up that wants people to use their voice instead of a keyboard.

According to Algérie 360, Zeghidour is 35 years old, has two children and leads one of the most prominent AI start-ups in Europe. Its models respond in 50 milliseconds, a speed Zeghidour describes as the fastest in the world.

From finance to artificial intelligence

He studied financial mathematics at Paris-Dauphine University, then completed an internship in quantitative finance. Faced with the flow of market information, he discovered machine learning and chose a specialised master's degree.

He then wrote his doctoral thesis at Facebook. He became the first doctoral student in its Paris laboratory. His subject was how babies learn language, and how to use this to design more efficient algorithms.

He worked on audio and speech recognition from the start of his career. He then joined Google Brain. With his future partner Olivier Teboul, he invented there the equivalent of large language models for audio.

This technology, which can generate voice as well as music, now sits behind Suno, ElevenLabs and the voice mode of OpenAI.

From Kyutai to Gradium

In 2023, he left Google to co-found Kyutai, a non-profit research laboratory based in Paris. Xavier Niel, Eric Schmidt and Rodolphe Saadé gave the project 300 million euros. The only instruction was to work on subjects with a high impact.

Six months later, the team unveiled Moshi, the first real-time voice conversation model to be made public. Unlike classic assistants, it listens and speaks at the same time, as a human does. Its work, published as open access, has since inspired many models around the world.

Requests from video game publishers and smartphone makers flooded in. This led him to change direction.

In September 2025, he co-founded Gradium with Olivier Teboul, Laurent Mazaré and Alexandre Défossez. The start-up was born as an offshoot of Kyutai, which is one of its main shareholders.

Gradium first raised 70 million dollars in December 2025, from investors including Eric Schmidt and Xavier Niel. In July 2026, Nvidia became a shareholder, and the fundraising reached 100 million dollars. The start-up opened an office in the San Francisco Bay Area soon after.

Voice in use today

With fewer than 25 staff, Gradium supplies its voice models to developers around the world. Renault already uses them, notably for voice transitions in podcast playlists listened to in cars.

The start-up plans to soon cover about 60 languages, including Arabic.

Zeghidour believes voice will become the main interface between humans and machines.

"I look forward to the moment when we can get rid of the keyboard," he said.

He believes the rise of AI agents, which people talk to in natural language, strengthens this change. He does, however, point out that voice models remain fragile on simple tasks. Saying a phone number or a number plate still causes problems for many of them.

The company's focus is reliability, speed and personalised voices, in any accent.