Home / Technology / Dessa: AI voice sounds eerily human
Dessa: AI voice sounds eerily human
24 Oct
Summary
- Block's Dessa AI trained solely on human conversations.
- Dessa uses speech-to-speech architecture for natural interaction.
- Dessa's audio output mimics human emotion and cadence.
Block, Inc. has introduced Dessa, an innovative AI model developed from the ground up using only human conversation data. This approach aims to create an AI that communicates with genuine emotion, nuance, and a natural cadence, distinguishing it from more robotic-sounding systems.
The core of Dessa's advancement lies in its speech-to-speech architecture. This technology allows the model to process and generate audio directly, facilitating simultaneous listening and speaking. This capability aims to eliminate the abrupt pauses common in other AI interactions.
Demonstrations of Dessa showcase its remarkably human-like voice and emotional range, with comparisons highlighting its superiority over other models like ChatGPT. The company has extended early access to a select group of AI safety researchers.
Dessa's development was driven by a desire for enhanced human-AI understanding and to address the safety concerns arising from AI's inability to perceive vocal emotion. The model can reportedly run locally on high-end MacBook Pro models.
