Hume brings emotionally intelligent voice AI models to production on IREN Cloud™

Background

Hume is an AI research company focused on voice and emotion, two areas that present unique technical challenges in AI development. Their work spans expressive text-to-speech, speech-language models, and the data and evaluation infrastructure needed to train them reliably. Building at this level requires sustained, high-performance compute backed by a team with the technical depth to support it.

The challenge

Training and iterating on expressive voice models is computationally intensive. Unlike text, voice carries layers of information including prosody, timing, emotion, and audio quality, all of which require large, high-quality datasets and continuous evaluation loops. Hume needed reliable access to GPU infrastructure that could support sustained training workloads, paired with an operations team that could respond quickly and knowledgeably when it mattered.

Solution and outcomes

To meet the compute demands of advanced voice model development, Hume partnered with IREN, a vertically integrated AI cloud provider delivering large-scale data centers and high-performance compute for AI training and inference. IREN's platform is underpinned by an expansive portfolio of grid-connected land and power in renewable-rich regions across the U.S. and Canada.

Hume leveraged IREN Cloud™, IREN's AI compute platform built to NVIDIA reference architecture with full non-blocking InfiniBand networking, high-performance storage, and the operational reliability that demanding AI workloads require.

IREN designs, builds, and operates its own data centers, giving its technical team direct visibility across the full infrastructure stack. When issues arise, they are handled by the same provider that built and runs the facility. For Hume, this means consistent performance across complex training workloads and a direct line to a team equipped to resolve issues fast.

"Working with IREN has given us the infrastructure reliability we need to move voice models from research into production. A lot of what we do depends on tight feedback loops between training and evaluation. Having infrastructure we can rely on, with a team that actually understands it, makes a real difference to how fast we can move."

— Nathan Leung, CTO, Hume AI

To bring their realistic, emotionally intelligent voice AI models to life, Hume utilizes IREN Cloud™ for reliable, high-performance compute infrastructure.

To learn how IREN Cloud™ is purpose-built for your AI workloads - Get in touch.