Thinking out loud
Blog.
-
Simpler, and still true: how we grade what Restate says
How a program checks that Restate's simpler sentences stay true, use words the listener knows, and fit the speaker's time, with numbers from 1,000 hours of speech.
Engineering Sep 20, 2026 · 16 min read
Dominik Roblek -
VoiceFrom delivers better translation accuracy than OpenAI's new GPT-Realtime-Translate
OpenAI shipped GPT-Realtime-Translate on May 8. We ran it through our head-to-head benchmark. VoiceFrom Interpret leads on translation accuracy by a wide margin.
News May 15, 2026 · 1 min read
Yahya Saleh -
Five platforms, one harness: a head-to-head live translation benchmark
We put five live translation platforms head-to-head. Here's the methodology behind the comparison, and where our system lands in the industry.
Engineering Apr 30, 2026 · 11 min read
Yahya Saleh -
What your audience actually waits for: Automating ear-voice span
A semantic latency metric for live speech translation that measures when meaning arrives, not when audio starts. We automate the 50-year-old gold standard of interpretation research.
Engineering Apr 19, 2026 · 9 min read
Yahya Saleh -
LLMs as translation judges: Inside GEMBA-MQM v2
The state-of-the-art translation quality metric: it lists the specific mistakes in a translation, not just how bad it is overall, without needing a gold-standard reference.
Engineering Apr 18, 2026 · 10 min read
Yahya Saleh
Nothing here yet. The lab is busy.