PolyAI Unveils Dialog-RSN-1: Audio-Native Model for Real-Time Calls with Lowest Latency

matthen2 · x · 2026-07-30

PolyAI introduces Dialog-RSN-1, a dialog model that directly perceives user audio, fusing turn-taking, ASR, function calling, and response into a single audio-native model. It is already handling live calls in production with the lowest latencies ever measured, enabling more fluid and intelligent conversations. The architecture overcomes the information bottleneck of cascaded systems by preserving audio signal and uncertainty, allowing the model to detect hesitation, emotions, accents, and speaking speed, and to interrupt or leave messages appropriately.

Related event: PolyAI Launches Dialog-RSN-1: End-to-End Native Audio Conversational Model(3 posts)→

Original post →

More from Apps

Apps channel →