Much more accurate voice transcription, steadier persona motion
Improvements
- Voice and phone conversations are transcribed by a new speech model with much higher accuracy, and it now gets the conversation's context to work from: what your agent is about, so words specific to your topic come through, and background noise or an unclear moment no longer turns into words your participant never said. What someone actually told you arrives in their own words, not an approximation of them.
- Hiding Perspective branding is clearer now. Turning on Remove Perspective branding also takes the "Perspective AI Agent" line under the message box with it, and the setting is honored in the design page's Agent view and the conversation preview, not just in the live conversation. We'd still love it if you told everyone about us anyway.
- Voice conversations now open with Murmur, and the persona sits a little smaller and closer to the message box so it shares the screen with the conversation instead of dominating it. Jelly and Weave are still one tap away.
Fixes
- The persona animation ran at whatever rate the participant's screen refreshed, so on a high-refresh-rate display it moved at double its intended speed and read as frantic while the agent was speaking. Every screen now shows the same, intended pace.
- Testing your agent before you share it now shows exactly what participants will see. The Agent view on the design page and the conversation preview were both falling back to the older two-screen text and voice experience instead of the single conversation view.