Block, the financial technology company led by Jack Dorsey, has introduced a new artificial intelligence model called Dessa, designed to hold natural conversations by learning entirely from human speech rather than text.
The company describes Dessa as a full-duplex speech-to-speech model, meaning it can listen and speak simultaneously, allowing it to respond, interrupt and adjust its delivery in ways that more closely resemble human conversations.
Dorsey, the co-founder of Twitter, announced the model on X on Friday, saying it was developed to make interactions with artificial intelligence more natural by capturing emotion, pauses and the rhythm of everyday speech.
“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,” Dorsey wrote.
“Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence,” he added.
According to Block, conventional voice AI systems often convert spoken words into text before generating responses, a process that can lose information conveyed through tone, timing and other features of speech.
Dessa is designed to process audio directly, enabling it to produce responses that reflect conversational cues such as laughter, hesitation, changes in tone and emotional expressions.
In demonstrations published alongside the announcement, Dessa engages in conversations about films, personal experiences and life advice. It also reacts to unexpected statements, adjusts its speaking volume when asked and interrupts naturally during exchanges.
In one recording, the model responds to an anniversary surprise with an audible gasp and an enthusiastic reaction. Other demonstrations show it laughing during a conversation and helping a speaker recall a film title.
Block also released a demonstration featuring Dessa speaking with OpenAI’s GPT-Live-1, another AI voice model, to illustrate differences in conversational timing, interruptions and responses.
See also: South African Presidency Shuts Down Website Over Hacking Fears
The company said Dessa was designed to recognise and reproduce elements of human speech, including emotion, humour, frustration and changes in vocal intensity. It argued that understanding these signals could improve how AI responds to people in sensitive situations.
“Imagine being in serious pain while a voice AI has no idea and just keeps going,” the company said, explaining its concerns about the limitations of existing voice interfaces.
Block said Dessa was trained from scratch using human speech, without text data or a pretrained language model. Its developers say the system can run locally on consumer hardware, including a MacBook Pro equipped with an M4 Max chip or a more powerful machine.
The company’s research team, led by Ragavan Thurairatnam, has presented the project as an effort to explore whether AI can develop conversational abilities and acquire knowledge directly from audio.
Beyond making voice assistants sound more natural, Block said the technology could improve AI safety by helping systems recognise when users are distressed, in pain or communicating urgency through their voices.
The company argued that voice-based systems that fail to recognise such cues could respond inappropriately in sensitive situations, particularly as AI becomes more capable and autonomous.
However, the model’s performance has not yet been independently established through publicly available benchmarks. Some demonstrations also reveal awkward responses, including instances in which the system misidentifies a speaker or takes a conversation in an unexpected direction.
As of Saturday, Block had not made Dessa generally available to the public. The company said it was inviting a small group of AI safety researchers to explore the model, while broader access remained limited.
The launch adds to growing efforts across the technology industry to develop voice-based AI systems that can interact with people in real time without the delays and limitations associated with conventional speech interfaces.
About the Author
Cecilia Attah is a tech analyst with a degree from Benue State University. She covers tech news and startups at TechRegard with a focus on how technology is transforming Africa and shaping the global landscape.
