Observed arrival · 2026-11-06
Speaking Machine Labs: Teaching machines when to speak
A voice-AI research lab working on models that can follow group conversations and judge when to speak, wait, or yield the floor.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Voice-AI researchers building agents for group conversations
- Worth noticing
- The lab's MP-Bench comparison reports 87.5% for a transcript-reading text model versus 32.6% for the best real-time voice agent.
Field notes
The work is split into Floor, a low-latency turn-taking model intended to run on a CPU; Chorus, which treats speakers as separate streams; and Agora, a simulated-meeting training and evaluation environment. The page says its models are scored on real multi-voice meeting audio with overlap, crosstalk, and poor microphones, and says it intends to publish benchmarks; several implementation details remain redacted.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
One card from the complete issue