RealFeedApp Dessa is trained fully on human speech without text in the pipeline
Dorsey launched Dessa, a locally trained full-duplex model capable of simultaneous speech and listening, powered by a single NVIDIA GPU and human speech data.
- Field
- applying language models
- What they did
- Block released a voice model named Dessa that runs locally on a single NVIDIA GPU and mimics human expressions like laughs and pauses using only conversational data.
- Why it matters
- This could be useful for creating more natural voice interfaces, as current models often sound unnatural due to the use of text tags to drive emotion in speech generation.
#voice model#llm#block#dessa#nvidia
