Gradium, an AI voice startup originating from the Kyutai lab, has emerged from stealth mode, securing a $70 million seed funding round. The company is focused on commercialising audio language models (ALMs) that aim to enhance voice interactions with machines. ALMs are designed to process, understand, and generate natural language using audio-text data, outperforming traditional large language models in voice AI tasks such as speech recognition and synthesis.
Gradium's technology promises ultra-low latency and more realistic AI voices, supporting natural and expressive conversations. The platform launches with support for English, French, German, Spanish, and Portuguese, targeting applications in gaming, customer care, language learning, healthcare, and AI agents. The company's integrated system is designed to master accuracy, ultra-low latency, natural conversational flow, and expressive synthesis.
Led by CEO Neil Zeghidour, a former Google DeepMind researcher, Gradium aims to redefine voice AI by addressing the limitations of existing systems. The company's approach involves a novel audio-language modelling technique that unifies generation, transcription, and dialogue within a single neural architecture. Gradium's early financial backing positions it to compete in the rapidly evolving voice AI landscape.
