Loading video...

Video Failed to Load

Go Home

Pipecat 1.16 adds PerceptionReader, something I'm really excited about. How someone speaks can matter just as much as what they say, and voice agents have mostly been missing that. We’ve been testing it with Google Decision 1, released yesterday. Stream it audio and you get a pretty accurate read...

17,381 views • 5 days ago •via X (Twitter)

9 Comments

Jon Taylor's profile picture
Jon Taylor5 days ago

Code is here:

Philipp Schmid's profile picture
Philipp Schmid4 days ago

Well done! And great example for the Google Decision 1 audio capability.

Pablo Giménez's profile picture
Pablo Giménez4 days ago

Sarcasm is the one I'd test first. In Spanish a lot of it lives only in the intonation, the words themselves are perfectly polite. Does it hold up outside English?

Neil Stoker ✨'s profile picture
Neil Stoker ✨4 days ago

Nice demo! ✨ Would be great to see metrics for "focused" and/or "confused" - those could be pretty useful for conversational strategies and allowed user activities

AJ's profile picture
AJ4 days ago

How are you acting on this sentiment information? I built a classifier for function_tool selection because LLMs accuracy at selecting tools is quite poor

yung algo's profile picture
yung algo4 days ago

very nice

Thor 雷神 ⚡️'s profile picture
Thor 雷神 ⚡️4 days ago

Oh, this is awesome. Thanks so much for sharing. 🔥

Shawn Wen's profile picture
Shawn Wen4 days ago

agreed. on phone calls the 'how' we need most is timing, so curious if these signals will feed smart turn too

Pete Haas's profile picture
Pete Haas4 days ago

Great example of using the Decisions API. I could imagine extending this to non-verbal cues like sighs. I.E. A sigh before "sure" changes what the agent should do next. Have you tried reading wordless turns?

Related Videos