Jack Dorsey’s Block unveils new voice AI model that talks like a real human

Former Twitter CEO Jack Dorsey unveiled a new speech model called Dessa as part of his company Block. In a post on X, Dorsey stated that voice is the most natural way to interact with intelligence and Dessa is the most authentic full-duplex speech model. The compact model can run on a laptop, making it truly portable. Dorsey expressed pride in the Dessa team for pushing the boundaries of what is possible.

Dessa is remarkable because it listens and talks simultaneously, with no transcription layer between audio input and output. It is trained entirely on real human conversations and can hold a real-time conversation without any text or pretrained models. Unlike other voice AI models that convert voice to text, Dessa captures nuances, emotions, and pauses, making interactions feel more natural and human-like.

The company claims that Dessa is faster than real-time and does not have any latency. It runs locally on a MacBook Pro M4 Max or higher. While the model is currently in research preview and behind a waitlist, there are demos available publicly. The company emphasizes that Dessa can gauge a user's feelings and emotions, which is crucial for creating a better user experience and preventing potential dangers.

Dessa represents an advancement in deep learning models, opening new possibilities for researchers to explore language learning and efficiency. It is a unique achievement, as Dessa learns language from hundreds of words, unlike most models that rely on massive amounts of text. The model also learns accents, pauses, breathing patterns, humor, sarcasm, inflection, emotion, and when to stop talking, making it highly adaptable and capable of generating realistic and nuanced conversations.