<?xml version="1.0" encoding="UTF-8"?><oembed><type>video</type><version>1.0</version><html>&lt;iframe src=&quot;https://www.loom.com/embed/874be69831d64553b1692b790198f099&quot; frameborder=&quot;0&quot; width=&quot;3320&quot; height=&quot;2490&quot; webkitallowfullscreen mozallowfullscreen allowfullscreen&gt;&lt;/iframe&gt;</html><height>2490</height><width>3320</width><provider_name>Loom</provider_name><provider_url>https://www.loom.com</provider_url><thumbnail_height>2490</thumbnail_height><thumbnail_width>3320</thumbnail_width><thumbnail_url>https://cdn.loom.com/sessions/thumbnails/874be69831d64553b1692b790198f099-a904f894ef5a7629.gif</thumbnail_url><duration>471.283333</duration><title>Voice Interactions Setup and Live Testing</title><description>This Loom explains how to set up and publish Voice Interactions for voice-based roleplay with an agent that talks back. It shows that you must connect a voice provider key first (supports 11Labs and Deepgram, and you can connect more than one), otherwise you can only save drafts and cannot pick a voice or publish. The editor lets you describe the scenario in plain English, which auto-fills the persona and situation, and you can enable a welcome card and ice breaker, select a voice, add skills, and attach the interaction to an agent. It also notes that testing uses your own provider key with no separate sandbox, and publishing checks for a connected key, name, filled persona and situation, and a selected voice; changes create versioned iterations.</description></oembed>