New Delhi [India], October 24 (ANI): Jack Dorsey-led financial technology company Block has unveiled Dessa, a new artificial intelligence (AI) voice model trained entirely on human conversations without using text or pretrained models, as per a company statement.
According to the company, Dessa is a full-duplex speech-to-speech AI model capable of listening and speaking simultaneously while recognising and expressing emotions, sarcasm, laughter, frustration and sadness through natural voice interactions.
Unlike conventional voice AI systems that convert speech into text, Dessa processes audio directly, preserving emotional cues and variations in tone that may 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.”
The company further highlighted the complexity of training AI models on speech, noting that spoken words contain significantly more data than written text and vary with each utterance.
“LLMs need an insane amount of text. Dessa learned language from hundreds of times fewer words (in fairness, LLMs also have to learn C++ and whatever the hell Angular is),” it noted.
Despite these challenges, Block said it successfully figured out “how to train Dessa on only one NVIDIA GB200 NVL72.”
According to the company, Dessa can independently determine what to say, when to respond and how to express itself, including its voice and conversational style.
“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,” it said.
The AI model can react naturally to humour by laughing mid-sentence, adjust its speaking volume when asked and respond spontaneously without being explicitly trained for these behaviours, the company claimed.
“A model that learned to talk from sound alone is new territory, and we want outside eyes on it. We’re giving a small group of AI safety researchers early access to find the risks we haven’t anticipated,” the release said. (ANI)
