{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/093da995b82f453ca4a3a47bb6bee741\" frameborder=\"0\" width=\"1920\" height=\"1440\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1440,"width":1920,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1440,"thumbnail_width":1920,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/093da995b82f453ca4a3a47bb6bee741-9f89f52d8765d395.gif","duration":140.863,"title":"Small Audio Action Model for Instant Tools","description":"This Loom introduces the Small Audio Action Model and explains why it can make voice assistants faster than the current ASR to text to LLM tool call pipeline. Derek Wang argues that typical assistants wait for you to finish speaking, adding latency, and that users do not own the intelligence since models are hosted on companies like Meta or Instinct. The new approach feeds speech directly into a small audio-action language model that triggers tool calls immediately when it hears a name, event, or other key word, and commits or uncommits actions quickly if it is wrong. He demonstrates this speed with an example conversation about meeting Josh and going hiking, and notes the model was fine tuned with high quality data created during the hackathon using River and Memorable."}