Block’s new voice model sounds eerily human
techcrunch ·
Block, Jack Dorsey’s fintech company, has unveiled Dessa, a new voice AI model trained entirely on human speech data. This full-duplex model can listen and speak simultaneously while running locally on consumer-grade hardware like the MacBook M4 Max. Unlike traditional systems that convert speech to text, Dessa preserves emotional nuances such as laughter, pauses, and crying directly within the audio generation process. The company claims this approach eliminates the unnatural cadence often found in current voice assistants by avoiding intermediate text pipelines. Block is currently seeking safety researchers to test the model for potential risks, such as social engineering or deception. Although Square already utilizes voice AI for ordering, Dessa aims to enable more sophisticated, cost-effective business and consumer experiences. Public access to the model remains restricted at this time.
techcrunch