Block unveils locally-run voice AI model
In brief
- Block launches Dessa, a voice AI model running locally on a MacBook M4 Max.
- The full-duplex model replicates emotions without intermediate text processing.
- Block invites safety researchers to test Dessa before a potential public release.
Block, the company founded by Jack Dorsey, has introduced Dessa, a voice AI model that runs locally on a MacBook M4 Max. The company claims the model, trained solely on speech data and using a single NVIDIA GPU, replicates human expressions such as laughter, pauses, and crying.
Dessa operates as a full-duplex model, enabling simultaneous listening and speaking, much like OpenAI’s GPT-Live released earlier this year. Unlike traditional models that rely on an intermediate text processing step, Block asserts that emotion is embedded directly into the model without external tags. The company shares examples of conversations where the bot adopts varied voices, rhythms, and emotions, such as sounding tired or emotional.
While Block specializes in financial solutions for small businesses, it aims to enhance its existing voice services, like those used by Square for order management, with this model. Though Dessa is not yet publicly available, Block is calling on safety researchers to test it for potential risks of deception by voice agents.
What it changes for you
- Finance & BankingBlock's voice solutions could reduce costs and improve customer experience for small businesses by using a local model.
This summary gives you the essentials. The full article, its figures and quotes are on AI News & Artificial Intelligence | TechCrunch.
Read the original articleAI-generated summary of the source, checked by rules. How we write · Report an error