Block has launched Dessa, a groundbreaking voice model that can listen and speak simultaneously, capturing authentic human expressions and emotions while running locally on a MacBook M4 Max or better.
- Dessa runs locally on consumer-grade GPUs, including MacBook M4 Max.
- It is the first full-duplex speech-to-speech model trained solely on human speech.
- Block invites safety researchers to test Dessa for potential misuse risks.
What happened
Block, the company founded by Jack Dorsey, has introduced Dessa, a new voice AI model that mimics human conversations with a high degree of emotional realism. This model can perform full-duplex interactions, meaning it can listen and speak simultaneously, unlike traditional voice AI that processes speech through text conversion.
Dessa was trained entirely on human speech using a single NVIDIA GPU and runs locally on devices like the MacBook M4 Max. Block demonstrated the model’s unique capacity to express emotions such as tiredness, joy, and vulnerability, creating a more natural and engaging conversational experience.
Why it matters
Current voice AI systems often reduce voice to text, stripping away emotional nuances crucial to human communication. Dessa’s speech-first approach preserves these subtle vocal cues, making interactions feel more authentic and empathetic.
Block’s ability to run Dessa on consumer-grade hardware lowers barriers to deploying sophisticated voice assistants across business and consumer applications. This innovation could significantly enhance voice-driven experiences in finance, retail, gaming, and beyond by delivering more natural and emotionally intelligent AI.
What to watch next
Block has not yet released Dessa for public use but is seeking input from safety researchers to identify risks, particularly concerning the potential for voice AI to deceive or manipulate human listeners. How the company addresses these concerns will be critical for responsible deployment.
Looking forward, Block may integrate Dessa more deeply with its existing financial and commerce platforms, potentially revolutionizing voice interfaces for small businesses. Market observers will be watching how this technology competes with other advanced speech models like OpenAI’s GPT-Live and ElevenLabs’ solutions.