Block's Dessa is a voice AI trained only on human speech, with no text at all
Block's research lab has previewed Dessa, a speech-to-speech model you can hold a live conversation with. It was trained from scratch on human conversation, without text data or pretrained models.
By ScoviAI Team
Key points
- Dessa is a research preview from a small AI lab at Block
- It's a full-duplex speech-to-speech model: it listens and talks at the same time
- Trained from scratch on human conversation, with no text and no pretrained models
- Block says it's the first AI for realistic live conversation trained only on speech
Block, the company behind Square and Cash App, has previewed Dessa, an AI you talk to out loud. What makes it unusual is how it learned: only from recordings of people talking, with no written text at all.
Dessa comes from a small AI research lab at Block founded by former members of Dessa, a Toronto deep-learning startup.
How most voice assistants work
Most voice AI today is built in layers:
- Your speech is turned into text.
- A text-based language model writes an answer.
- The answer is turned back into speech.
Each step loses something. Text doesn't carry tone, hesitation, laughter or the way people overlap in a real conversation.
What's different about Dessa
- It's a speech-to-speech model: sound in, sound out, with no text step in between.
- It's full-duplex, meaning it can listen and speak at the same time, like people do.
- Block says it was trained from scratch on human conversation, with no text data and no pretrained models.
Block calls it the first AI you can hold a realistic live conversation with that was trained only on human speech. That claim comes from Block itself; there are no independent tests yet.
Why train on speech alone?
The idea is that a model that has only ever heard people talk could pick up the parts of conversation that text-based systems miss: timing, emotion and natural back-and-forth. The trade-off is knowledge. Text is where most of the world's facts live, so a speech-only model may be better at sounding human than at answering detailed questions.
What it means for you
Dessa is a research preview, not a product you can download into an app today. But it points to where voice assistants are heading: less like dictation, more like talking to a person. For voice tools you can use now, see our guide to ElevenLabs alternatives.
Sources
ScoviAI Team is the ScoviAI newsroom. We write every article ourselves from the original announcements and link our sources. Spotted a mistake? Tell us and we'll fix it.
ADVERTISEMENT
