Block Dessa Voice Model Brings Human-Like Speech to Local AI
Jack Dorsey’s Block has released a new voice model out of the company’s research lab, and it sounds strangely human. The Block Dessa voice model runs locally and is capable of expressions like laughs, pauses, and cries. Block says it is trained fully on human speech and a single NVIDIA GPU rack.
Dessa is a full-duplex model, meaning it can talk and listen at the same time, just like the GPT-Live model OpenAI released earlier this year. Typically, speech models are paired with a large language model for knowledge retrieval. Block said that Dessa’s knowledge stems from the conversational data it has.
In a post on X, Jack Dorsey wrote: “We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.”
Why Block Says Current Voice AI Falls Short
Block argues that current models can sound unnatural and annoying because they have text in the pipeline that drives speech generation. Many models use voice tags to drive emotion in speech generation. Dorsey’s company claims that emotion is built into its model, though it is not clear whether that emotion is being steered by some kind of tags.
“Interacting with current voice AI can just plain suck. It turns your voice into text and throws away an incredible amount of information text alone can’t carry. The most powerful AIs in the world can’t hear how you feel. We think that’s ridiculous, leads to a worse experience, and is potentially dangerous,” the company said in a blog post.
The core criticism is that converting speech to text strips away tone, pacing, and emotional cues. By keeping the pipeline in speech, the Block Dessa voice model aims to preserve those signals instead of flattening them.
What the Block Dessa Voice Model Sounds Like in Practice
The company’s researchers posted multiple examples of conversations with Dessa in which the agent had different voices, cadences, and emotions. In one conversation, the bot said it was tired from work, and at that point it sounded tired.
In another example, the bot was talking about the person making it comfortable while sounding emotional and almost breaking down. A further demo shows a Block team member playing a guessing game with the voice assistant. The model guesses the first one right straight away, and the second doesn’t take long either.
Ragavan Thurairatnam, who shared a demo on X, described the release as a milestone: “We made the first AI model you can have a realistic conversation with that was trained entirely on speech! Dessa is a full-duplex speech-to-speech model that you can run on consumer GPUs. This is a GPT3 moment for speech.”
Full-Duplex Conversation Changes the Feel
Because Dessa can listen and speak simultaneously, it can handle interruptions and back-and-forth exchanges more naturally than turn-based systems. That design choice is central to why the Block Dessa voice model feels less like a scripted assistant and more like a participant in a conversation.
Local Hardware and Limited Access
Block said the model can run locally on a consumer-grade computer such as a MacBook M4 Max or better. However, there is no public access to Dessa at the moment, and Block has put out a call for safety researchers to test the model. This is likely to test cases where a voice agent could trick people on the other end.
Running on-device also means conversations do not have to leave the machine, which could matter for privacy-sensitive use cases. Still, the limited release means most people cannot try the Block Dessa voice model yet.
Why the Block Dessa Voice Model Matters for Business
Block is in the business of finance and selling its solutions to small businesses. Square already uses voice AI to handle ordering. With its own voice model that can run on-device, Block could power many more business and consumer experiences at a lower cost.
For small businesses, a locally running voice model could reduce reliance on cloud inference and cut operating expenses. That fits Block’s broader push to sell practical tools to merchants rather than research demos alone.
The Road Ahead for Dessa
The safety researcher call signals that Block wants outside scrutiny before opening access. Voice agents that sound this human raise real questions about deception, consent, and trust on the other end of a call. How Block answers those questions will shape whether the Block Dessa voice model becomes a widely used tool or a controlled experiment.
