· 3 items
arXiv ↗
DiffuPlex: Accelerating Full-Duplex Spoken Dialog Models via Rolling…
- What
- DiffuPlex introduces a rolling masked diffusion framework for fast full-duplex speech models.
- Why
- It eliminates sequential compute bottlenecks in simultaneous listening-speaking AI agent interfaces.
- Impact
- Crucial for building low-latency real-time voice agents.
arXiv ↗
EgoVoice: Proactive Spoken Assistance from Egocentric Multimodal Stre…
- What
- EgoVoice enables proactive spoken assistance using egocentric multimodal streams.
- Why
- It shifts voice agents from reactive command-followers to proactive context-aware assistants.
- Impact
- Directly applicable pattern for advanced voice agent UX design.
arXiv ↗
OnTrack: Real-Time Monitoring and Intervention in LLM Agent Trajector…
- What
- OnTrack offers real-time monitoring and intervention for autonomous LLM agent trajectories.
- Why
- It prevents costly and unsafe irreversible actions during multi-turn autonomous execution.
- Impact
- Improves reliability for production voice and action agents.