Emotive voice AI startup Hume launches new EVI 3 model with rapid custom voice creation

New York-based AI startup Hume has unveiled its latest Empathic Voice Interface (EVI) conversational AI model, EVI 3 (pronounced “Evee” Three, just like the Pokémon character), focusing on every little thing from powering buyer assist techniques and well being teaching to immersive storytelling and digital companionship.

EVI 3 lets customers create their very own voices by speaking to the mannequin (it’s voice-to-voice/speech-to-speech), and goals to set a brand new commonplace for naturalness, expressiveness, and “empathy” in response to Hume — that’s, how customers understand the mannequin’s understanding of their feelings and its skill to reflect or regulate its personal responses, by way of tone and phrase alternative.

Designed for companies, builders, and creators, EVI 3 expands on Hume’s earlier voice fashions by providing extra subtle customization, quicker responses, and enhanced emotional understanding.

Particular person customers can work together with it immediately by Hume’s live demo on its website and iOS app, however developer entry by Hume’s proprietary utility programming interface (API) is alleged to be made out there in “the approaching weeks,” as a blog post from the company states.

At that time, builders will have the ability to embed EVI 3 into their very own customer support techniques, artistic tasks, or digital assistants — for a worth (see under).

My very own utilization of the demo allowed me to create a brand new, customized artificial voice in seconds primarily based on qualities I described to it — a mixture of heat and assured, and a masculine tone. Chatting with it felt extra naturalistic and straightforward than different AI fashions and positively the inventory voices from legacy tech leaders such Apple with Siri and Amazon with Alexa.

What builders and companies ought to learn about EVI 3

Hume’s EVI 3 is constructed for a spread of makes use of—from customer support and in-app interactions to content material creation in audiobooks and gaming.

It permits customers to specify exact character traits, vocal qualities, emotional tone, and dialog matters.

This implies it could possibly produce something from a heat, empathetic information to a unusual, mischievous narrator—right down to requests like “a squeaky mouse whispering urgently in a French accent about its scheme to steal cheese from the kitchen.”

EVI 3’s core power lies in its skill to combine emotional intelligence immediately into voice-based experiences.

In contrast to conventional chatbots or voice assistants that rely closely on scripted or text-based interactions, EVI 3 adapts to how folks naturally communicate — selecting up on pitch, prosody, pauses, and vocal bursts to create extra participating, humanlike conversations.

Nevertheless, one massive characteristic Hume’s fashions at the moment lack — and which is obtainable by rivals open supply and proprietary, similar to ElevenLabs — is voice cloning, or the speedy replication of a person’s or different voice, similar to an organization CEO.

But Hume has indicated it should add such a functionality to its Octave text-to-speech mannequin, as it’s famous as “coming quickly” on Hume’s web site, and prior reporting by yours really on the corporate discovered it should enable customers to duplicate voices from as little as 5 seconds of audio.

Hume has acknowledged it’s prioritizing safeguards and moral concerns earlier than making this characteristic broadly out there. At the moment, this cloning functionality shouldn’t be out there in EVI itself, with Hume emphasizing versatile voice customization as an alternative.

Inner benchmarks present customers want EVI 3 to OpenAI’s GPT-4o voice mannequin

In response to Hume’s personal assessments with 1,720 customers, EVI 3 was most well-liked over OpenAI’s GPT-4o in each class evaluated: naturalness, expressiveness, empathy, interruption dealing with, response velocity, audio high quality, voice emotion/type modulation on request, and emotion understanding on request (the “on request” options are lined in “instruction following” seen under).

It additionally normally bested Google’s Gemini mannequin household and the brand new open supply AI model firm Sesame from former Oculus co-creator Brendan Iribe.

It additionally boasts decrease latency (~300 milliseconds), strong multilingual assist (English and Spanish, with extra languages coming), and successfully limitless customized voices. As Hume writes on its web site (see screenshot instantly under):

Key capabilities embrace:

• Prosody era and expressive text-to-speech with modulation.
• Interruptibility, enabling dynamic conversational stream.
• In-conversation voice customizability, so customers can regulate talking type in actual time.
• API-ready structure (coming quickly), so builders can combine EVI 3 immediately into apps and companies.

Pricing and developer entry

Hume presents versatile, usage-based pricing throughout its EVI, Octave TTS, and Expression Measurement APIs.

Whereas EVI 3’s particular API pricing has not been introduced but (marked as TBA), the sample suggests it will likely be usage-based, with enterprise reductions out there for giant deployments.

For reference, EVI 2 is priced at $0.072 per minute — 30% decrease than its predecessor, EVI 1 ($0.102/minute).

For creators and builders working with text-to-speech tasks, Hume’s Octave TTS plans vary from a free tier (10,000 characters of speech, ~10 minutes of audio) to enterprise-level plans. Right here’s the breakdown:

• Free: 10,000 characters, limitless customized voices, $0/month
• Starter: 30,000 characters (~half-hour), 20 tasks, $3/month
• Creator: 100,000 characters (~100 minutes), 1,000 tasks, usage-based overage ($0.20/1,000 characters), $10/month
• Professional: 500,000 characters (~500 minutes), 3,000 tasks, $0.15/1,000 additional, $50/month
• Scale: 2,000,000 characters (~2,000 minutes), 10,000 tasks, $0.13/1,000 additional, $150/month
• Enterprise: 10,000,000 characters (~10,000 minutes), 20,000 tasks, $0.10/1,000 additional, $900/month
• Enterprise: Customized pricing and limitless utilization

For builders engaged on real-time voice interactions or emotional evaluation, Hume additionally presents a Pay as You Go plan with $20 in free credit and no upfront dedication. Excessive-volume enterprise clients can go for a devoted Enterprise plan that includes dataset licenses, on-premises options, customized integrations, and superior assist.

Hume’s historical past of emotive AI voice fashions

Based in 2021 by Alan Cowen, a former researcher at Google DeepMind, Hume goals to bridge the hole between human emotional nuance and AI interplay.

The corporate skilled its fashions on an expansive dataset drawn from a whole lot of 1000’s of members worldwide—capturing not simply speech and textual content, but in addition vocal bursts and facial expressions.

“Emotional intelligence consists of the power to deduce intentions and preferences from conduct. That’s the very core of what AI interfaces try to realize,” Cowen instructed VentureBeat. Hume’s mission is to make AI interfaces extra responsive, humanlike, and in the end extra helpful—whether or not that’s serving to a buyer navigate an app or narrating a narrative with simply the appropriate mix of drama and humor.

In early 2024, the corporate launched EVI 2, which supplied 40% decrease latency and 30% lowered pricing in comparison with EVI 1, alongside new options like dynamic voice customization and in-conversation type prompts.

February 2025 noticed the debut of Octave, a text-to-speech engine for content material creators able to adjusting feelings on the sentence degree with textual content prompts.

With EVI 3 now out there for hands-on exploration and full API entry simply across the nook, Hume hopes to permit builders and creators to reimagine what’s doable with voice AI.