Gradium

Gradium's Voice Design lets developers generate a new synthetic voice from a written prompt, specifying attributes like accent, age, gender, and pace, and receive a usable voice in seconds through the Gradium API or Studio. The feature is aimed at teams building voice agents, apps, and other spoken interfaces who want a specific character on demand rather than picking from a fixed catalog of stock voices. It ships free at launch, which matters because voice casting is often the slowest, most subjective step in shipping an agent, and iterating on a text prompt is far cheaper than commissioning or cloning talent. Gradium is a voice AI platform founded by Neil Zeghidour, Laurent Mazaré, Olivier Teboul, and Alexandre Défossez, the co-founders of Kyutai, a research lab with peer-reviewed work on audio language models . The company raised an unusually large $70 million seed round to commercialize the voice-AI algorithms that its founders, ex-DeepMind and Meta researchers, say underpin today's transcription and synthesis tech , giving Voice Design real weight as an early product from a team with deep provenance in generative audio. The launch fits Gradium's broader thesis that audio language models, trained on datasets that pair audio with descriptive text, can learn the complex relationships between sound and language well enough to be steered by natural prompts. For founders and operators shipping voice features, that translates to faster experimentation on tone and persona without swapping vendors or waiting on studio time, and for investors it is a concrete first look at how the team plans to turn its research into a developer product.