Universal 3.5 Pro Streaming adds Context Carryover (Full Transcript)

[00:00:00] Speaker 1: Today we launched Universal 3.5 Pro Streaming, our latest flagship speech-to-text model. The feature that I'm most excited about is Context Carryover. Here's the problem that it solves. Most transcription models hear each moment on its own, with no memory of what came before it. So when someone gives a short answer like C, there's nothing for the model to disambiguate it with, and it just guesses. Context Carryover fixes that automatically. The model remembers the last few things that were said, and uses them as context for whatever comes next. Let me use a practical example to show you what I mean. I can compare Universal 3.5 Pro with our older speech-to-text model here. And if I ask a question like, Do you speak Spanish? C. Did you arrive here by land, air, or sea? C. Do you like option A, B, or C? C. So as you can see, the model takes each question as context for the answer, and transcribes it with a different form of C each time. And this is all done automatically by default on Universal 3.5 Pro Streaming. And so give it a try today, especially for voice agents, if you haven't already. I'm so excited to see what you build with Universal 3.5 Pro Streaming. Bye.