Block, the financial technology company led by Jack Dorsey, has released a new voice model called Dessa. The model was built by the company’s research lab and shared by Dorsey on X.
Dessa is designed to sound very human. It can laugh, pause, and even cry during a conversation.
Block said the model was trained fully on human speech. The training was done using a single NVIDIA GPU rack.
How Dessa Works
Dessa is a full-duplex model. This means it can talk and listen at the same time, much like a person does in a normal conversation.
OpenAI released a similar type of model, called GPT-Live, earlier this year. Both models aim to make live voice conversations feel more natural.
Most speech models are paired with a large language model to look up knowledge. Block said Dessa works differently, with its knowledge coming from the conversational data it was trained on.
Block said many current voice models sound unnatural because text sits in the middle of the process. The text in the pipeline is what drives the speech the model produces.
Some models, such as ElevenLabs v4 and Fish Audio Drama 3, use voice tags to add emotion to speech. Block claims emotion is built directly into Dessa. It is not yet clear whether some kind of tags are still steering that emotion.
In a blog post, the company said, “Interacting with current voice AI can just plain suck.” It added, “The most powerful AIs in the world can’t hear how you feel.”
Block called that situation “ridiculous” and said it leads to a worse experience. The company also described it as “potentially dangerous.”
Examples and Next Steps
Block researchers shared several sample conversations with Dessa. In these clips, the AI used different voices, speaking speeds, and emotions.
In one example, the bot said it was tired from work, and its voice sounded tired. In another, it spoke about the person making it feel comfortable and sounded emotional, almost breaking down.
A third clip showed a Block team member playing a game with the voice assistant. The examples were shared to show how the model sounds in real conversations.
Block’s main business is finance, including selling tools to small businesses. Its Square unit already uses voice AI to take orders.
Block said Dessa can run locally on a consumer-grade computer, such as a MacBook M4 Max or better. TechCrunch reported that an on-device model could let the company power more business and consumer services at a lower cost.
For now, Dessa is not available to the public. Block has put out a call for safety researchers to test the model, likely to study cases where a voice agent could trick people on the other end of a call.
