Hume AI is moving voice assistants closer to conversations that feel emotionally aware. The company has raised $50 million in a Series B funding round and unveiled the Empathic Voice Interface, or EVI, a flagship product designed to respond not only to words, but also to how those words are spoken.
The result is a voice AI that invites comparison with Samantha from the movie "Her". Hume AI is not presenting EVI as a digital partner, but the technology points toward a new category of assistants that may feel more personal, more responsive and more difficult to treat as ordinary software.
A new round backs empathic AI
EQT Ventures led Hume AI’s $50 million Series B round. The funding also included participation from Union Square Ventures, Nat Friedman & Daniel Gross, Metaplanet, Northwell Holdings, Comcast Ventures and LG Technology Ventures.
The financing arrives alongside the launch of EVI, which Hume AI describes as a voice interface with emotional intelligence. In plain terms, the company is trying to build an assistant that can listen for signals beyond spoken content and use those signals to shape the conversation.
That is the core distinction from a standard voice assistant. The source frames EVI as an Alexa with emotional intelligence: a system that can process the user’s words while also paying attention to cues such as tone, pitch, volume and timing.
How EVI listens differently
EVI is powered by what Hume calls the Empathic Large Language Model (eLLM), a new form of multimodal generative AI. According to the company, the model has been trained on data from millions of human interactions.
The interface uses tone and volume to understand when a user has finished speaking. That matters because real conversations are not built from clean turns, fixed commands or perfect pauses. People trail off, interrupt themselves, shift their pitch and reveal meaning through delivery.
Hume AI says EVI reads mood and optimizes its responses over time "for happiness and satisfaction." The system is designed to recognize natural variations in pitch and tone that can carry meaning beyond the literal words. It also responds with low latency, which can make the exchange feel closer to a live conversation.
Several details define the product’s promise:
- It listens for vocal signals, not only text content.
- It uses tone and volume to judge conversational timing.
- It is built around mood recognition and adaptive responses.
- It relies on the Empathic Large Language Model (eLLM).
- It is intended to feel fast enough for natural dialogue.
A demo version of EVI is already live, powered by Claude 3 Haiku for text generation. Hume AI plans to make the technology publicly available to developers in April, with a waiting list now open.
Why the comparison to "Her" is hard to avoid
The movie "Her", starring Joaquin Phoenix and Scarlett Johansson, centers on Theodore, a lonely man who forms a deep bond with Samantha, an AI voice that appears to understand him more deeply over time. Their friendship eventually becomes romantic.
EVI is not described as a system meant to recreate that story. Still, the comparison is useful because it captures why emotionally responsive voice AI feels different from a normal chatbot. A convincing voice, fast replies and mood-aware responses can make software feel socially present.
The source suggests that Hume’s EVI could make at least the friendship part of such a bond more realistic. That is a narrow but important point. The technology does not need to simulate romance to create a sense of closeness; it only needs to make users feel heard, understood and responded to in a personal way.
That may be valuable in many applications, but it also changes the level of trust users may place in a system. If an AI can recognize hesitation, sadness, excitement or comfort through the voice, users may share more with it than they would with a conventional tool.
The privacy and ethics questions
Hume AI says EVI is not meant to simulate a relationship. The interface emphasizes that it is an AI system, not a real person. Even so, an emotionally attuned AI could become a place where users disclose private feelings, secrets and vulnerable thoughts.
That creates a sharper privacy issue than ordinary voice commands. The sensitive material is not only what users say, but also how they sound while saying it. Mood, tone and emotional state become part of the interaction.
The ethical questions are equally direct. Should a company encourage emotional connections between humans and machines at all? If users develop an unhealthy dependency on AI partners, who is responsible? The source notes that this can already happen with text chatbots alone.
Hume AI aims to address such issues through the Hume Initiative, a nonprofit that brings together AI researchers, ethicists, social scientists and legal experts to craft ethical guidelines for empathic AI.
What EVI signals about voice AI
EVI points to a shift in what voice assistants may become. Earlier assistants were mainly command systems: ask a question, set a reminder, request a task. Hume AI’s approach suggests a future in which voice assistants are judged by how well they understand emotional context, not just by whether they return the right answer.
That makes the technology more powerful and more complicated. A system that adapts to mood could feel useful, supportive and easy to talk to. The same qualities could also make users more likely to treat it as a companion.
A full "Her"-like relationship remains a distant scenario, according to the source. But Hume AI’s EVI shows that emotionally aware AI voice assistants are no longer only a science fiction idea. They are becoming products, demos and developer tools, backed by serious funding and aimed at real-world applications.