The iRTE2026 conference highlights the shift in AI voice interaction from single-turn Q&A to full-duplex continuous communication. Agora, as an infrastructure provider, showcased its underlying capabilities supporting models like OpenAI and Gemini, aiming to lower development barriers by reusing network and interaction components. The event primarily reflects industry trends and vendor marketing, lacking specific technical breakthroughs or parameter details.
Full-duplex Communication
全双工通信
A communication mode allowing simultaneous transmission and reception by both parties.
Curated Intelligence
The iRTE2026 conference highlights real-time intelligent interaction, marking a shift in conversational AI from single-turn generation to continuous participation in real exchanges. Leading products like OpenAI's GPT-Realtime and Google's Gemini Live are driving this evolution through full-duplex, multimodal, and tool-calling capabilities. While reflecting industry focus on low-latency voice agent infrastructure, the event lacks specific new technical parameters or breakthrough releases.
Block has released Dessa, a full-duplex AI voice model that claims to generate native emotional expressions like laughter and sobbing without relying on text intermediaries. Trained on human speech data using NVIDIA GPU racks, the model supports local execution on M4 Max or higher Macs and is currently available only to security researchers for testing deception risks.
Block has released Dessa, a full-duplex voice model capable of emotional expression and local execution on a single GPU. It relies on conversational data rather than external LLM retrieval, aiming to enable low-cost on-device business applications.
Block has unveiled Dessa, an AI speech model trained entirely on human speech from scratch without any text data. By processing audio directly, Dessa captures emotions, sarcasm, and pauses in real time while supporting full-duplex communication; it currently runs locally on a MacBook Pro M4 Max and is not yet publicly available.
Startup Dessa has released a full-duplex speech model trained entirely from scratch on human audio conversations, bypassing traditional text-based pretraining. By processing raw acoustic signals, the model preserves paralinguistic cues like emotion and cadence, aiming for more natural real-time interactions.
SpaceXAI has released Grok Voice 3.0, claiming it as the world's most intelligent speech-to-speech model with a top score of 86.9 on the Artificial Analysis leaderboard. The model achieves sub-second latency and outperforms competitors in enterprise benchmarks like τ-Voice and EVA while maintaining lower operational costs. It is already deployed for Tesla vehicle interactions and Starlink customer support, with API pricing remaining unchanged at $0.08 per minute.
openJiuwen has launched WorkSwarm, enabling full-duplex multimodal agent interactions similar to a phone call. This solution allows users to interrupt or share visuals while the AI executes complex background tasks, addressing latency issues in traditional turn-based dialogue. The core innovation lies in the dual-rhythm coordination between real-time models and the Core Agent, with native affinity for Ascend computing power.
OpenAI introduces GPT-Live-1 in the API to enable natural, full-duplex voice conversations. The model features improved instruction following, custom voices, and telephony support.