Dessa - Full-Duplex Speech Model
What it is
Dessa is a research preview of a full-duplex speech-to-speech model developed by Block. It is trained exclusively on human conversation audio rather than text, allowing it to process and generate speech in real-time with natural emotional nuance, pauses, and backchanneling.
The problem it solves
Current voice AI systems convert speech to text, discarding critical non-verbal cues like emotion, tone, and timing, resulting in stilted, slow, and unnatural user experiences.
What you could use it for
- Real-time voice assistants that require natural, low-latency conversation flow
- Research into speech-only language model training and efficiency
- Voice interfaces that need to detect and respond to emotional tone and sarcasm
Why it matters
Establishes a new category of AI interaction that preserves the full information content of human speech, potentially leading to more natural and emotionally intelligent voice interfaces that do not rely on lossy text transcription.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.