Jack Dorsey’s Block has unveiled a new voice model named Dessa, developed in the company’s research lab. This model runs locally and is designed to sound remarkably human, incorporating expressions such as laughs, pauses, and cries. Block claims that Dessa is trained entirely on human speech using a single NVIDIA GPU.

Features of Dessa

Dessa is a full-duplex model, meaning it can simultaneously talk and listen, similar to the GPT-Live model released by OpenAI earlier this year. Unlike typical speech models that rely on large language models for knowledge retrieval, Dessa derives its knowledge from conversational data.

“

"We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence," Block stated in a blog post.

Comparison with Existing Models

Block highlighted that current voice models often sound unnatural and can be frustrating to interact with. Many models, such as ElevenLabs v4 and Fish Audio Drama 3, utilize voice tags to convey emotion in speech generation. However, Block asserts that emotion is inherently built into Dessa’s model.