NEW DELHI: Jack Dorsey-led financial technology company Block has unveiled Dessa, an artificial intelligence (AI) voice model trained entirely on human conversation, without using text or pretrained models.
According to the company, Dessa is a full-duplex speech-to-speech model that can listen and speak at the same time. It is designed to recognise and express emotions, including sarcasm, laughter, frustration and sadness, during real-time conversations.
Unlike conventional voice AI systems that convert speech into text before processing it, Dessa works directly with audio. This allows it to retain emotional cues, changes in tone and other nuances that can be lost during text conversion.
“We trained Dessa from scratch on nothing but human conversation. Sounds! No text, no pretrained models. It learned to hold a real-time conversation simply by listening,” Block said.
The company highlighted the challenges of training AI on speech, noting that spoken language carries more information than written text and varies with every utterance.
“LLMs need an insane amount of text. Dessa learned language from hundreds of times fewer words,” Block said, adding that it had worked out how to train the model using a single NVIDIA GB200 NVL72 system.
Block said Dessa can determine what to say, when to respond and how to deliver its replies, including its voice and conversational style. The model can laugh in response to humour, adjust its speaking volume when asked and respond spontaneously, according to the company.
“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,” the company said.
Block is also opening the model to a small group of AI safety researchers for early access, seeking external scrutiny to identify potential risks.
“A model that learned to talk from sound alone is new territory, and we want outside eyes on it,” the company said.
