Radar ·
Dessa: full-duplex speech-to-speech model for consumer GPUs, no primary source
A post announces Dessa, a full-duplex speech-to-speech model that, according to its author, was trained only on speech and runs on consumer GPUs. No primary source is linked and the claims have not been verified.
According to the post dated 23 October 2026, Dessa is a full-duplex speech-to-speech model, meaning it handles speech input and output at the same time. The author states that it was trained entirely on speech and that it runs on consumer GPUs. The text provides no primary source, benchmarks or architecture details, so these claims have not been verified.
The item matters to anyone evaluating conversational voice models that can be hosted locally. Before considering any use, consult the original post and look for a technical article, repository or model card from the authors. Check the license, hardware requirements and any results measured by third parties.
Original source: Dessa: modelo speech-to-speech full-duplex para GPUs de consumo