About nyra labs
Building speech intelligence for the whole signal
The central insight
Speech is more than words
Speech technology was built to produce clean transcripts. But hesitation, repetition, timing, rhythm, pronunciation, effort, and silence all carry information.
Working with people recovering speech after neurological injury showed us how much is lost when those details disappear.
Why this lab exists
Built from the hardest speech problems
nyra labs grew out of nyra health, where speech technology is used in neurological rehabilitation. That work brings us into contact with speech that conventional systems were never designed to understand.
It creates a direct feedback loop: difficult cases become research questions, new models return to real use, and what fails in practice comes back to the lab as the next problem to solve.
Our research programme
One programme, built layer by layer
- 01
Preserve
Capture what happened, including the pauses, repetitions, repairs, and vocal sounds a clean transcript removes.
- 02
Measure
Locate every word, sound, speaker, and event at the level where it actually occurs.
- 03
Represent
Separate content, speaker, prosody, and acoustics into spaces we can inspect and understand.
- 04
Understand
Build models that can learn efficiently from everything the speech signal carries.
Work with us
Build models that listen closer
Join us to build open speech models grounded in the places where every detail matters.