01
Acoustic-first
We start with the raw signal — breath, pace, timbre — before a single word is guessed. Recognition that holds up in noisy rooms, soft asides, and everything in between.
Private beta
Coming soon
Every voice,
fully understood.
Auris is building voice recognition that hears what people mean — not just what they say. Accent-native, context-aware, and fast enough to disappear.
01The vision
Beyond transcription
Speech is more than words.
Most systems hear a sequence of sounds and return text. We think that's only the beginning.
Accent, emphasis, hesitation, intent — the things that make us human are the things today's voice models throw away. Auris is built to keep them.
- Tone
- Intent
- Emotion
- Accent
- Context
140+
Languages in a single model, not a patchwork.
<40ms
Target end-to-end latency for live conversation.
On·device
Private by design. Fast enough to run where you are.
v1.0 targets · benchmarks published at launch
02The technology
A new voice stack
Built from the signal up.
A new architecture for a new era of voice. Three principles guide every layer we build.
02
Context that compounds
Auris carries the thread of a conversation. Names, acronyms, speaker changes, and the thing you almost said stay in view — turn after turn.
03
Fast enough to disappear
Streaming inference and on-device processing make voice feel less like a command and more like a conversation. No wait. No cloud round-trip required.
The stack
Raw audio
Acoustic model
Context engine
Intent
Your product
03Early access
Be first to hear it.
Auris is coming soon. Join the waitlist for early access, launch updates, and the occasional look under the hood.