Research notes
Notes from the work.
Position papers and design notes on real-time, speech-native conversation. These describe directions and reasoning. When we have results we are prepared to defend, they will get notes of their own.
03—October 2026
11 min read
A latency budget for live speech
Where the time goes between a caller's mouth and the agent's voice, which costs are physics and which are design, and why a live speech system has to stream at every stage.
Read the note02—October 2026
11 min read
Turn-taking is a model problem, not a timer problem
What people actually do when they hand the floor back and forth, why silence thresholds cannot reproduce it, and what a model that handles turn-taking natively has to learn.
Read the note01—October 2026
11 min read
The case for speech-to-speech
Why the transcribe, think, synthesize pipeline cannot be fixed with faster parts, what a speech-native model has to learn instead, and where the hard problems are.
Read the note