July 2025 Summaries

Hume AI and SambaNova have partnered to integrate emotionally intelligent voice capabilities into SambaNova's language models, addressing the demand for advanced text-to-speech (TTS) solutions within their existing AI infrastructure. This collaboration leverages SambaNova's ultra-fast inference capabilities, achieving response times between 100 ms and 300 ms, which are crucial for real-time voice applications. Hume's Empathic Voice Interface (EVI 3) now natively supports SambaNova-hosted models like Llama and DeepSeek, providing developers with options that balance speed and cost-effectiveness while delivering expressive, emotionally aware voice interactions. The integration offers enterprise customers seamless, reliable voice AI solutions, combining SambaNova's robust AI infrastructure with Hume's expertise in emotional speech, thereby enhancing user experience in various applications.

Hume AI has collaborated with Groq to integrate emotionally intelligent voice capabilities into Groq-hosted language models, launching with the Kimi K2 multimodal assistant. This partnership combines Groq's LPU™ Inference Engine's ultra-low latency processing with Hume's Empathic Voice Interface (EVI) to produce human-like conversations with sub-300ms speech-to-speech latency and emotional understanding. EVI interprets and responds to emotional cues in speech, enhancing communication authenticity by dynamically adjusting prosody, rhythm, and inflection. Groq's deterministic performance ensures real-time applications can maintain natural conversational flow without delays, making it a reliable choice for developers looking to build scalable, expressive voice applications. The integration supports seamless development on Hume's platform, allowing teams to focus on application logic while leveraging Groq's advanced capabilities for creating engaging, lifelike AI interactions.

Hume AI and Cerebras have collaborated to integrate Hume's Empathic Voice Interface (EVI) with Cerebras-hosted language models, showcasing the Qwen-3-235b-a22b reasoning model. This partnership combines Cerebras' ultra-fast inference infrastructure with Hume's emotionally intelligent voice technology to create voice agents capable of complex problem-solving and natural, emotionally expressive speech. By leveraging Cerebras' performance-optimized infrastructure, the Qwen-3-235b-a22b model can process complex reasoning tasks in real-time, facilitating seamless conversational flow without typical latency issues. This integration allows developers to create sophisticated voice applications with emotional intelligence, suitable for educational tutoring, customer service, and professional assistance, enabling voice agents to engage in nuanced and natural conversations.

Vapi's platform, in collaboration with Hume AI, provides developers with a streamlined solution for creating conversational voice agents in phone calling applications, integrating Hume's Octave TTS for real-time expressive text-to-speech at a low latency of 150ms and competitive pricing of 2¢/minute. This integration not only offers significant performance and cost improvements but also allows developers to focus on application logic by simplifying the complex orchestration of speech recognition, language, and voice synthesis models. Vapi's commitment to optimizing speed, reliability, and affordability has led to a 41% cost reduction and a 66% latency improvement, making it an attractive option for developers focusing on realistic, expressive speech and cost efficiency. The platform's enhanced expressiveness, with emotionally aware voice outputs, has been widely adopted by developers across various industries, such as healthcare, customer service, and education, who value the combination of affordability and emotional expressiveness. Alan Cowen, CEO of Hume AI, emphasizes that the partnership with Vapi demonstrates that developers can achieve both emotional intelligence and practical constraints like cost and speed without compromise, making the EVI integration a compelling choice for building advanced voice AI applications.

EVI 3, the latest empathic voice interface from Hume, offers groundbreaking features in speech-language modeling, enabling expressive speech with any voice without fine-tuning, and introducing hyperrealistic voice cloning with less than 30 seconds of audio. This model surpasses traditional text-to-speech systems by integrating language and voice processing into a single, faster, and higher quality speech-to-speech model, trained on trillions of text tokens and millions of speech hours. EVI 3's interoperability allows seamless integration with other popular large language models like Claude 4 and Gemini 2.5, enhancing its capabilities in AI applications such as assistants and VR characters. The platform supports over 200,000 user-designed voices and offers efficient communication with other systems, ensuring low latency and customizable speech interactions. EVI 3 is accessible at competitive pricing, and developers can explore its features through demos, playgrounds, and comprehensive documentation, setting the stage for innovative AI-driven projects.