arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2304.04472cs.CL
收藏

建模说话者-听话者交互以预测反馈信道

Modeling Speaker-Listener Interaction for Backchannel Prediction

University of Stuttgart(斯图加特大学)

Daniel Ortega, Sarina Meyer, Antje Schweitzer, Ngoc Thang Vu

AI总结:

本文提出一种基于神经网络的声学反馈信道分类器,通过提取说话者语音特征并编码说话者-听话者动态交互,在Switchboard和GECO数据集上结合双方行为嵌入实现了最佳F1分数的预测性能。

摘要:

我们展示了关于反馈信道建模的最新发现,其新颖动机源于英语中最小回应词“Yeah”和“Uh-huh”的规范用法及其在德语中的对应词,以及对说话者-听话者交互进行编码的效果。反馈信道理论强调听话者在对话过程中的主动和持续作用、其对说话者后续谈话的影响,以及由此产生的动态说话者-听话者交互。因此,我们提出了一种基于神经网络的声学反馈信道分类器,用于处理最小回应词,通过从说话者语音中提取声学特征,捕捉并模仿听话者的反馈信道行为,并对说话者-听话者交互进行编码。我们在Switchboard和GECO数据集上的实验结果表明,在几乎所有测试场景中,说话者或听话者的行为嵌入都有助于模型做出更准确的反馈信道预测。更重要的是,一种恰当的交互编码策略,即结合说话者和听话者嵌入,使得模型在两个数据集上的F1分数均达到最佳性能。

英文摘要:

We present our latest findings on backchannel modeling novelly motivated by the canonical use of the minimal responses Yeah and Uh-huh in English and their correspondent tokens in German, and the effect of encoding the speaker-listener interaction. Backchanneling theories emphasize the active and continuous role of the listener in the course of the conversation, their effects on the speaker's subsequent talk, and the consequent dynamic speaker-listener interaction. Therefore, we propose a neural-based acoustic backchannel classifier on minimal responses by processing acoustic features from the speaker speech, capturing and imitating listeners' backchanneling behavior, and encoding speaker-listener interaction. Our experimental results on the Switchboard and GECO datasets reveal that in almost all tested scenarios the speaker or listener behavior embeddings help the model make more accurate backchannel predictions. More importantly, a proper interaction encoding strategy, i.e., combining the speaker and listener embeddings, leads to the best performance on both datasets in terms of F1-score.

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。

↑