Block's new AI model Dessa was trained entirely on human speech to catch emotions text-based models miss

Block has introduced an audio-based AI model named Dessa that processes human speech directly without transcribing it into text first. By training exclusively on conversational audio, the system aims to identify emotional nuances like sarcasm and pauses that text-based models often overlook. This approach enables real-time responsiveness and can operate locally on consumer hardware. While currently not available to the public, the development represents a shift toward more natural, context-aware vocal interactions between humans and AI.