Lightning talk: training a small audio model

Turn detection is a major challenge for conversational voice AI. I’ll talk about training a new open source, open code, open data “smart turn detection model.” I’ll touch on model architectures for audio processing tasks, the latest developments in this space from the big labs (OpenAI Realtime API and Gemini Multimodal Live API) and startups, creating data sets, evaluating training runs, deploying a small model like this at scale in production, and how anyone can get involved.