Hume rolls out browser voice app powered by EVI 2 and Claude

Most frontier voice AI demos force you through app store downloads, waitlists, and sign-up flows. Hume AI just cut through all of that friction.

The emotional AI research lab launched a new browser-based voice application that lets anyone chat via microphone instantly, without creating an account or logging in.

EVI 2 meets Claude 3.5 Haiku

Under the hood, Hume is running a hybrid setup that plays to its specific strengths.

Speech generation and vocal personality are powered by Hume's proprietary speech-language model, EVI 2. Meanwhile, general reasoning and conversational text are handled by supplementary models, notably Anthropic's Claude 3.5 Haiku.

The web app packs several specialized personas tailored to different tasks. Users can switch between a quick-answers bot for factual inquiries, a philosophical persona for deeper discussions, and a storytelling bot for creative narratives.

Early testing shows real promise and rough edges

According to initial hands-on testing by getcoai, the app shows both impressive flashes and clear growing pains.

The quick-answers persona delivered solid factual accuracy, handling straightforward queries without hallucinating or losing track.

Creative tasks proved trickier. The storytelling assistant occasionally repeated dialogue lines and suffered from unexpected voice shifts mid-conversation.

Testers also noted some content overlap across different personas, suggesting the distinct personalities still need tighter guardrails.

Why it matters

OpenAI has already planted its flag in conversational audio with ChatGPT's Advanced Voice Mode, but Hume's browser-first, zero-login approach gives it an immediate distribution edge.

For regular users, it lowers the barrier to entry to essentially zero. You can test expressive, low-latency voice interaction right now without handing over an email address or paying a monthly subscription.

We think Hume's architectural choice is the smartest part of this release. Instead of burning capital trying to out-train frontier labs on general reasoning, pairing specialized speech models with efficient external engines like Claude 3.5 Haiku offers a viable blueprint for challenger labs.

That said, conversational stability remains the biggest hurdle. If voice consistency and persona drift aren't ironed out quickly, casual users won't stick around once the novelty wears off.