nyra labs

Research

How we build speech models that keep what others erase — the methods, the open benchmark, the data engine, and the models themselves.

Featured

All research

Longform transcription with conditional continuationResearch · Jun 25, 2026
CrisperWhisper 2.0Model · Jun 24, 2026
Measuring verbatimnessBenchmark · Jun 24, 2026
Closing the verbatim data gapDataset · Jun 23, 2026
Faster inference and mitigating hallucinations in CrisperWhisper 2.0Research · Jun 22, 2026
How we unlocked multilingual style-controlled ASR at scaleResearch · Jun 21, 2026
Turning emergent cross-attention into a precise alignerResearch · Jun 20, 2026