What changed
Block, the company behind Square and Cash App, has unveiled an AI speech model called Dessa. Dessa is the first AI model trained entirely on human speech, with no text data at all. Instead of converting audio to text first, Dessa processes speech directly as audio, allowing it to pick up on emotions, sarcasm, laughter, and pauses in real time. The model supports full-duplex communication, enabling it to listen and talk simultaneously. It runs locally on a MacBook Pro M4 Max and was trained on a single NVIDIA GB200 NVL72.
What this means in practice
Dessa is designed to improve human-AI communication by capturing emotional cues that text-based models often miss. Block developed Dessa out of curiosity, a desire for better human-AI interaction, and safety concerns, as current voice AI systems ignore emotional cues. The model's ability to process speech directly as audio may lead to more natural and emotionally responsive interactions in applications such as customer service, virtual assistants, and mental health support tools.
Limitations and availability
Dessa is not publicly available yet. The team is first looking for AI security researchers to test the model for unexpected risks. Anyone interested can sign up via a form. The model's local operation on a MacBook Pro M4 Max suggests it may have specific hardware requirements, though details on broader compatibility are not provided. The supplied announcement does not specify the exact number of parameters or the model's licensing conditions.
