Jack Dorsey’s Block has built a new voice AI model called Dessa, and the company is presenting it as a speech system that sounds less polished and more like an awkward new acquaintance than typical voice assistants. The model is the work of a Dessa team led by Ragavan Thurairatnam, who shared a sample conversation in which Dessa mostly holds a natural back-and-forth with a speaker but initially mistakes the person for someone named Ronaldo.
The Dessa website offers additional examples of the model in conversation. In one exchange, Dessa discusses Christopher Nolan’s films and veers into an unexpected tangent about Bryan Singer’s Superman films. In another, a person describes taking Dessa on a surprise trip to Paris, and the site says Dessa responds with what sounds like tearful gratitude. A further sample pairs Dessa with OpenAI’s GPT-Live-1 voice model, and Dessa comes across as the more natural of the two; clips from some of the conversations appear in a video.
On its website, the Dessa team describes the system as a true speech-to-speech model whose entire knowledge base came from human speech. The team says it can run locally on a MacBook Pro M4 Max or better. The team gives three reasons for building it: curiosity, which it also calls fun; better understanding between humans and AI; and safety.
The team elaborated on those motivations. It said babies learn to talk without ever seeing a written word, and the team wanted to see whether a model could do the same, which it can, though the effort is very hard. The team argued that speech is the most natural interface to any intelligence and that current voice AI wastes it by flattening speech into text and failing to keep up with human timing.
The safety rationale centers on emotion. The team said AI keeps gaining power and independence while remaining nearly deaf to the emotion in human voices, and offered the example of a voice AI continuing on while a person is in serious pain. Dessa is described as a step toward fixing that problem.
The Dessa team says it will give a small group of AI safety researchers early access to the model. The release arrives as US technology companies continue to push voice interfaces into consumer and developer products, a market where natural-sounding speech and the ability to read a speaker’s emotional state are active areas of competition.
More AI news from TechManNews.
