PACE: A Playback-Aligned Context Engine for LLM-Based Full-Duplex Voice Dialogue

2026-08-07

AI Summary

This work addresses the “Generated Context Misalignment” (GCM) problem in large language model–driven full-duplex voice conversations, where asynchrony between text generation and audio playback causes contextual inconsistencies. To resolve this, the authors propose PACE, a middleware that anchors dialogue context to the client-side playback boundary and dynamically truncates unplayed assistant utterances upon user interruption, re-anchoring context without requiring modifications to black-box speech models. PACE ensures consistent behavior across heterogeneous runtime environments by integrating end-to-end browser-based audio processing, real-time playback progress tracking, and dynamic context adjustment. Evaluated on a newly introduced GCM-Bench, PACE improves referential anchoring accuracy from 25.0% to 96.3%, while maintaining high-quality, low-latency responses under interruptions in the Full-Duplex-Bench v1 benchmark.

Abstract

LLM-based full-duplex voice services allow users to speak while the assistant is responding. Because servers can generate output and advance dialogue state faster than clients can play it, subsequent user speech may be interpreted based on content the user never heard. We call this failure Generative Context Mis-anchoring (GCM). To address GCM issues, we present PACE, a provider-independent middleware layer that anchors model-facing context to the client playback boundary, a system-observable proxy for what the user could have heard. After an interruption, PACE repairs this context to exclude assistant content that never reached playback, while preserving low-latency generation across heterogeneous voice runtimes. We implement PACE's audio-only projection path end to end in a browser-based realtime voice assistant using a black-box speech model, without modifying the model service. We also construct GCM-Bench, a new controlled benchmark dataset of 108 playback-relative referent-anchoring cases. On GCM-Bench, PACE raises Referent Anchoring Accuracy from 25.0% to 96.3% over a cancellation-only baseline. On 200 Full-Duplex-Bench v1 interruption samples, it preserves interruption response quality. These results show that grounding model-facing context in actual playback is a practical way to maintain consistency in full-duplex voice dialogue.

Problem

• full-duplex voice dialogue
• Generative Context Mis-anchoring
• context anchoring
• playback alignment
• LLM-based voice assistant

Innovation

• full-duplex voice dialogue
• context alignment
• playback boundary
• generative context mis-anchoring
• referent anchoring