Home / Technology / AI Speaks: Dessa Learns Language From Sound Alone
AI Speaks: Dessa Learns Language From Sound Alone
25 Oct
Summary
- Dessa processes audio directly, preserving emotional cues.
- The AI model can laugh, adjust volume, and respond spontaneously.
- Researchers will examine potential risks of the new AI.
Block has introduced Dessa, a groundbreaking AI voice model that bypasses text entirely, learning to converse directly from human speech.
This full-duplex AI processes audio inputs in real-time, allowing it to simultaneously listen and speak while accurately interpreting and conveying emotions like sadness and laughter. Dessa's unique training method, which relies exclusively on sounds and avoids pre-trained models or text conversion, preserves the natural nuances of human speech.
The company developed Dessa to interact more authentically, enabling it to react spontaneously to humor, adjust its vocal volume, and exhibit a natural conversational flow. To ensure responsible development, Block is providing early access to a small cohort of AI safety researchers for risk assessment.
