Dessa

Block announced 'Dessa', a full-duplex voice conversation model trained exclusively on human conversation audio, as a research preview.

• •Block announced the full-duplex voice conversation model 'Dessa' on October 23 (US time).
• •The model was trained only on human conversation audio and can respond in real-time with backchannels.
• •It was released as a research preview and achieves natural dialogue similar to human-to-human conversation.

Block announced a research preview of 'Dessa', a full-duplex voice model trained exclusively on human conversational voice without using text or pre-trained models.

• •It is a full-duplex voice model trained exclusively on human conversational voice without using text or pre-trained models.
• •It is characterized by its ability to directly hear and express backchannels and emotional intonation.
• •It was trained on a single NVIDIA rack and can be run locally on a MacBook Pro.

Block announced an AI model named Dessa trained entirely on human conversation data and shared chat logs to demonstrate its capabilities.

• •Block released an AI model called Dessa, trained from scratch using only human conversation data.
• •The company shared chat logs with humans to showcase the model's conversational abilities.
• •Some of the shared conversations occasionally appear awkward or uncanny.

An article introduces the research preview of 'Dessa', a full-duplex speech model designed to learn speaking by listening, mimicking human language acquisition.

• •'Dessa' is released as a research preview of a full-duplex speech model.
• •The design philosophy mimics the human process of acquiring language skills through listening.
• •It presents a technical approach aiming for a more natural conversational experience than traditional speech models.