Hume AI brings expressive speech to SambaNova-powered language models | Hume Blog | Hume AI

Hume AI has partnered with SambaNova to bring expressive, emotionally intelligent voice to SambaNova-hosted language models. The collaboration addresses a key need identified by SambaNova's extensive customer base, who have been seeking SambaNova-hosted TTS solutions to complement their existing LLM infrastructure, showcasing models like Llama-4-Maverick-17B-128E-Instruct and DeepSeek-R1-Distill-Llama-70B with seamless voice experiences within their trusted AI platform.

Key partnership highlights:

• Native integration of SambaNova-hosted TTS and LLMs with EVI 3 platform
• Cost-effective frontier LLM options including DeepSeek-R1-Distill-Llama-70B for budget-conscious developers
• Ultra-fast inference optimized for real-time voice interactions

Ultra-fast inference meets expressive voice

SambaNova delivers high-performance AI infrastructure optimized for large language models. When Hume was evaluating additional LLM providers for EVI 3, SambaNova's ultra-fast inference capabilities, achieving response times between 100 ms and 300 ms and meeting the 200 ms conversational latency benchmark, along with its diverse model offerings, made it an attractive partner for expanding developer choice.

This partnership brings together SambaNova's ultra-fast frontier LLM inference with Hume's real-time speech-to-speech model, giving developers more options for building voice applications that combine speed, cost-effectiveness, and emotional intelligence.

Real-time emotional speech with EVI 3 + SambaNova

At the heart of the partnership is Hume's Empathic Voice Interface, a speech-to-speech AI model that understands user emotional intonation and word emphasis and responds with expressive, contextually appropriate voice. SambaNova's fast inference capabilities make it particularly well-suited for real-time voice applications where response latency directly impacts user experience.

EVI 3 now supports SambaNova-hosted Llama and DeepSeek models natively, providing customers with supplemental LLM options that maintain the performance standards they expect from SambaNova infrastructure while adding emotional intelligence to their voice applications.

Why developers choose SambaNova Llama models in EVI 3:

• Fast inference performance: SambaNova's infrastructure supports Hume’s EVI 3 and frontier LLMs with 100 to 300 ms response times, meeting the 200 ms conversational latency benchmark for natural, real-time voice
• Frontier intelligence: SambaNova-hosted LLMs unlock frontier intelligence without sacrificing on conversational latency
• Cost-effective options: Including Llama and DeepSeek models that provide excellent value for budget-conscious projects
• Enterprise reliability: Built on SambaNova's RDU (Reconfigurable Data Unit) infrastructure, a hardware architecture trusted by its existing customer base for delivering consistent, high-performance AI inference at scale
• Seamless integration: Native support within EVI 3 requires no additional orchestration

Together, SambaNova and Hume provide enterprise customers with a complete voice AI solution that combines fast, reliable inference with sophisticated emotional intelligence.

"SambaNova's fast inference and diverse model options, including cost-effective DeepSeek variants, give our developers the flexibility to balance performance and budget while maintaining real-time voice responsiveness." — Alan Cowen, CEO of Hume AI

Start building with voice AI powered by Hume and SambaNova

SambaNova-hosted Llama models are now available within EVI 3, giving enterprise customers and developers access to fast, cost-effective language models optimized for real-time voice interactions. Whether you're building internal tools, customer-facing applications, or exploring new voice AI use cases, this integration provides enterprise-grade reliability with emotional intelligence.

Explore SambaNova Llama models in EVI 3 →
Get in touch to discuss enterprise voice AI solutions →