Hi Hugo, it would be great if you could discuss turn detection and interruption in Voice AI systems in your upcoming videos.

I know this group tends to lean more toward voice AI over telephony. However, I thought some would still find this project I put together to help me at next weekend's CalHack, where I will be assisting contestants with their implementation questions and judging the projects. This does not demonstrate the actual use case, but gives an idea of how to use a super small device to interact with voice AI I was at a PhysicalAI meetup a couple of weeks ago and spoke to a lot of folks there. I realized the deep thirst for more physical AI, so that got me wanting to make more examples to help folks get started. youtube.com/watch?v=B_A6xlY_zQs&feature=youtu.be Working on getting the code core changes added, then will open source it. In a similar vein, Sean Dubios and I were chatting recently, and I was glad to hear his magical phone is now in stores https://mrchristmas.com/products/santas-magical-telephone. I am not marketing or getting anything out of this. I am just happy to see physical embodiments of voice AI, and Sean is a hero of mine.

When I compare total pricing for a Voice AI in Livekit/Pipecat vs. VAPI/Retell I don't see the point of going with Livekit or Pipecat as it looks like the prices I would pay for SST, LLM and TTS in VAPI/Retell are better compared to what I would pay directly to e.g. Deepgram, OpenAI and Elevenlabs via Livekit/Pipecat. Even when self-hosted. Am I wrong?

Our agency started out as a general agency. We offered every solution possible. It was easy to get clients to start with... But it wasn't sustainable. There were two issues: 1. we became a jack of all trades, master of none 2. it was really hard to scale So we decided to niche down on Voice AI Voice AI in itself it still a very broad sector of AI, but it gave us a chance to master a skill. I decided to make a video outlining why we chose Voice AI above anything else.