Block, Inc. unveils Dessa, an AI voice model trained on human conversation

Block, Inc. has introduced Dessa, an AI voice model designed to replicate human conversation with emotional nuance and natural cadence. Unlike conventional voice models, Dessa processes and generates audio directly, enabling more realistic interactions. The model is currently in early access for AI safety researchers.

Tarun Mottlia

Editor, Lazyfounder

Published Oct 24, 20263 min read

Block, Inc. unveils Dessa, an AI voice model trained on human conversation — Image: © rootstock via Shutterstock via source

30 SEC SUMMARY

Block, Inc. releases Dessa, an AI voice model trained exclusively on human conversation.
Dessa generates audio directly using speech-to-speech architecture, enabling natural cadence and emotional nuance.
The model can speak and listen simultaneously, reducing awkward pauses in interactions.
Early access granted to AI safety researchers; runs locally on high-end MacBook Pro devices.

KEY HIGHLIGHTS

Dessa is trained exclusively on human conversation, enabling emotional nuance and natural cadence.
Uses speech-to-speech architecture to generate audio directly, unlike text-based voice models.
Can speak and listen simultaneously, reducing interaction delays.
Runs locally on MacBook Pro M4 Max or better devices.
Early access limited to AI safety researchers; no public release timeline.
Block, Inc. aims to improve human-AI emotional understanding and address safety concerns.

How Dessa works

Dessa is trained exclusively on human conversation, allowing it to mimic emotional nuance, pauses, and natural speech patterns. According to Block, Inc., the model’s speech-to-speech architecture enables it to generate audio directly, bypassing the text-based intermediaries used by traditional voice models.

The model’s duplex architecture allows it to speak and listen simultaneously, reducing the awkward pauses typical in AI interactions. This design aims to create a more fluid and human-like dialogue experience.

Capabilities and limitations

Dessa’s voices are described as uncannily human, with the ability to convey emotion and cadence in ways that distinguish it from competitors like ChatGPT, which often sound robotic. A demonstration quoted in reports highlights its naturalistic flow: the model responded, "Man, I’m doing good, breh. I ain’t gonna even lie. I’m tired."

The model runs locally on devices with MacBook Pro M4 Max or better specifications, ensuring low-latency performance. However, its capabilities are currently limited to research access, with no public release timeline disclosed.

Purpose and early access

Block, Inc. developed Dessa to explore human-AI interaction, improve emotional understanding in AI, and address safety concerns related to voice models. The company has granted early access to a select group of AI safety researchers to evaluate its potential and risks.

Dessa’s focus on emotional nuance reflects a growing trend in AI development, where models are increasingly designed to interpret and replicate human-like context, not just text or speech.

What this means

Lazyfounder analysis — our interpretation, not reported fact.

Dessa represents a shift in AI voice models toward more natural, emotionally aware interactions. For founders and operators in AI, this signals a growing emphasis on human-like dialogue capabilities, which could redefine user expectations for voice-based AI applications. The local execution on consumer hardware also hints at a broader trend of democratizing advanced AI tools, reducing reliance on cloud-based processing. However, the limited early access underscores the importance of safety and ethical considerations in deploying such technology, particularly in applications requiring emotional sensitivity.

Key takeaways

AI voice models are evolving toward more natural, emotionally nuanced interactions, which could set new benchmarks for user experience.
Local execution of advanced models on consumer hardware may reduce costs and latency for AI-powered applications.
The focus on safety and ethical considerations in early access reflects broader industry caution around emotionally aware AI.
Founders in voice-based AI, customer support, or conversational interfaces may need to reassess their technology stacks to compete with models like Dessa.

Sources

Gizmodo · 2026-10-24
Check Out the Many Uncanny, Human-ish Voices From Block’s New AI Model

This story is an original summary drafted with AI by Lazyfounder from the reporting listed above and checked by automated validation. Facts are attributed to their original publishers; sections marked as analysis are Lazyfounder's. Where a source is in another language, facts were machine-translated and quotations are reported, not reproduced. Read the original coverage via the links, and see our AI policy and corrections policy.

About the author