Neil Zeghidour is co-founder and chief executive of Gradium, which builds speech models and infrastructure for real-time voice applications. His research includes Google’s SoundStream and AudioLM and Kyutai’s Moshi, a conversational model capable of listening and speaking simultaneously.
On this page
From audio compression to voice agents
Gradium’s launch announcement identifies Zeghidour’s research background at Meta and Google DeepMind.
He was the first listed author of SoundStream, a neural codec for compressing speech, music, and other audio. He subsequently co-authored AudioLM, which generates coherent speech and music using discrete audio representations.
In September 2025, Zeghidour co-founded Gradium with Olivier Teboul, Laurent Mazaré, and Alexandre Défossez. The company launched publicly on December 2 with $70 million in seed funding and a focus on voice infrastructure, including streaming speech recognition, synthesis, and voice cloning.
What natural voice AI actually requires
Full-duplex speech: Human conversations include interruptions, overlapping voices, and brief acknowledgments. Moshi can listen and respond concurrently, avoiding the rigid conversational turns that cause many voice assistants to stumble.
Production-ready conversational agents: Natural speech alone cannot replace dependable tool use, safety controls, observability, and personalization. In his Europe talk, Zeghidour acknowledges that conventional speech-recognition, language-model, and speech-synthesis pipelines remain practical to inspect and deploy, and argues that integrated speech systems must match their reliability, intelligence, and personalization before replacing them.
Application-level latency: External searches and tool calls can delay conversations more severely than speech synthesis. His demonstrated approach keeps an assistant speaking naturally while information is retrieved, then incorporates the result without an awkward pause. It improves conversational continuity rather than speeding up the underlying retrieval.
Paralinguistic understanding: Hesitation, discomfort, tone, and other vocal signals convey information that transcripts discard. Zeghidour argues that speech models must be trained to recognize and act on these cues, not merely receive audio inputs.
Gradium Phonon:Gradium’s on-device text-to-speech model, presented by Zeghidour, runs on a smartphone CPU, making speech synthesis less dependent on cloud inference and allowing TTS processing to stay local. This does not establish that recognition, reasoning, tools, or an entire interaction remain on the device. His stated goal is practical consumer-scale voice AI without prohibitive synthesis API costs; licensing and other application costs still matter.