arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2605.20356cs.CLcs.AIcs.SD

全双工语音对话模型中的同步与轮流机制

Synchronization and Turn-Taking in Full-Duplex Speech Dialogue Models

ASAPP Inc.(ASAPP公司);Departamento de Computación, FCEyN, Universidad de Buenos Aires(计算机系,福克雷斯-恩分校,布宜诺斯艾利斯大学)

Pablo Riera, Pablo Brusco, Cristina Kuo, Marcelo Sancinetti, S. R. K. Branavan

AI总结:

本文研究了全双工语音对话模型如何通过内部表示的同步协调交互,并发现噪声条件下同步性下降,内部状态编码了提前的轮流预测信息。

摘要:

全双工语音对话模型(SDMs)能够同时听和说,使交互动态更接近人类对话。受人类沟通中神经耦合的启发,我们研究了此类模型在交互过程中如何协调其内部表示。我们模拟了两个预训练Moshi模型在受控条件下的全双工对话,操纵信道噪声和解码偏置。通过跨时间滞后计算中心核对齐(CKA)来测量同步性,同时利用因果LSTM模型从说话者和倾听者角度探测提前的轮流提示信号。我们发现无噪声条件下同步性较强,接近零滞后,随着噪声增加而下降,并展示了内部状态编码了支持提前轮流预测的信息。

英文摘要:

Full-duplex spoken dialogue models (SDMs) can listen and speak simultaneously, enabling interaction dynamics closer to human conversation than turn-based systems. Inspired by neural coupling in human communication, we study how such models coordinate their internal representations during interaction. We simulate full-duplex dialogues between two instances of the pretrained \textit{Moshi} model under controlled conditions, manipulating channel noise and decoding bias. Synchronization is measured using Centered Kernel Alignment (CKA) across temporal lags, while anticipatory turn-taking cues are probed from delayed internal activations using causal LSTM models, from both speaker and listener perspectives. We find strong representational synchronization under no noise conditions, peaking near zero lag and degrading with noise, and we show that internal states encode anticipatory information that supports turn-taking prediction ahead of time.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑