Skip to main content
AI Voice AgentsOctober 30, 20265 min read

Block Unveils Dessa, a Full-Duplex Speech Model Trained Only on Speech

Block’s AI lab has introduced Dessa, a new full-duplex speech model that runs locally on a MacBook Pro M4 Max. Trained solely on human conversation audio, Dessa marks a major advance in AI voice agents, promising better and faster communication tools for sales and customer-service teams.

By BrightLaunchIQ, Intelligence Team
Block Unveils Dessa, a Full-Duplex Speech Model Trained Only on Speech editorial illustration
BrightLaunchIQ AI Industry News

Block recently launched Dessa, a groundbreaking full-duplex speech-to-speech model designed to operate using only speech training data. Unlike many AI voice models trained on a mix of data types, Dessa focuses exclusively on audio from human conversations. This helps create more natural and fluid communication between users and AI systems.

One impressive feature of Dessa is that it runs locally on a MacBook Pro M4 Max. This means organizations can use the model without heavy cloud processing, improving privacy and reducing latency. For sales and customer-service teams, this technology can lead to faster responses and more natural interactions with clients or customers.

Full-duplex technology enables simultaneous speaking and listening, which is closer to how human conversations work. Instead of waiting for a person to finish speaking before responding, Dessa can process and reply in real time, making voice-enabled agents feel more like real conversations. This improves user experience and efficiency for small businesses that rely on phone support or AI assistants.

By training Dessa only on speech data, Block’s AI lab has focused the model development on the nuances of human voice and conversation. This choice could lead to better understanding of tone, pacing, and speech patterns when compared to models trained on broader data sets that include text or other non-speech inputs.

For small businesses and organizations using voice technology in sales or customer service, Dessa represents a tool that could reduce overhead costs and improve customer satisfaction. Running the model locally also means less dependence on internet connectivity and cloud platforms, which can be vital for remote or resource-limited settings.

Looking ahead, it will be important to watch how Block further develops Dessa and whether similar models start to appear across different devices and platforms. The ability to run advanced speech models locally could shift how many companies deploy AI voice agents in their day-to-day operations, enabling faster and more secure communication.

Overall, Dessa is a notable step forward in AI voice technology that emphasizes real human speech and local processing. As AI voice assistants become more common in business settings, advancements like Dessa will likely shape how teams interact with customers and streamline their communication processes.

Keep reading

BrightLaunchIQ Intelligence Team