Brutschi Robin, Wang Rui, Kolbe Michaela, Weiss Kerrin, Lohmeyer Quentin, Meboldt Mirko
D-MAVT, ETH Zurich, Leonhardstrasse, Zurich, 8092, Zurich, Switzerland.
Simulation Center USZ, Universitätsspital Zürich, Huttenstrasse 46, Zurich, 8091, Zurich, Switzerland.
Adv Simul (Lond). 2024 Oct 9;9(1):42. doi: 10.1186/s41077-024-00315-1.
Debriefings are central to effective learning in simulation-based medical education. However, educators often face challenges when conducting debriefings, which are further compounded by the lack of empirically derived knowledge on optimal debriefing processes. The goal of this study was to explore the technical feasibility of audio-based speaker diarization for automatically, objectively, and reliably measuring debriefing interaction patterns among debriefers and participants. Additionally, it aimed to investigate the ability to automatically create statistical analyses and visualizations, such as sociograms, solely from the audio recordings of debriefings among debriefers and participants.
We used a microphone to record the audio of debriefings conducted during simulation-based team training with third-year medical students. The debriefings were led by two healthcare simulation instructors. We processed the recorded audio file using speaker diarization machine learning algorithms and validated the results manually to showcase its accuracy. We selected two debriefings to compare the speaker diarization results between different sessions, aiming to demonstrate similarities and differences in interaction patterns.
Ten debriefings were analyzed, each lasting about 30 min. After data processing, the recorded data enabled speaker diarization, which in turn facilitated the automatic creation of visualized interaction patterns, such as sociograms. The findings and data visualizations demonstrated the technical feasibility of implementing audio-based visualizations of interaction patterns, with an average accuracy of 97.78%.We further analyzed two different debriefing cases to uncover similarities and differences between the sessions. By quantifying the response rate from participants, we were able to determine and quantify the level of interaction patterns triggered by instructors in each debriefing session. In one session, the debriefers triggered 28% of the feedback from students, while in the other session, this percentage increased to 36%.
Our results indicate that speaker diarization technology can be applied accurately and automatically to provide visualizations of debriefing interactions. This application can be beneficial for the development of simulation educator faculty. These visualizations can support instructors in facilitating and assessing debriefing sessions, ultimately enhancing learning outcomes in simulation-based healthcare education.
总结汇报是基于模拟的医学教育中有效学习的核心。然而，教育工作者在进行总结汇报时常常面临挑战，而关于最佳总结汇报流程的实证性知识的缺乏又进一步加剧了这些挑战。本研究的目的是探索基于音频的说话人识别技术在自动、客观且可靠地测量总结汇报者与参与者之间的互动模式方面的技术可行性。此外，其旨在研究仅从总结汇报者与参与者之间的总结汇报音频记录自动创建统计分析和可视化图表（如社会关系图）的能力。
我们使用麦克风记录了在与三年级医学生进行的基于模拟的团队培训期间进行的总结汇报的音频。总结汇报由两名医疗模拟教员主持。我们使用说话人识别机器学习算法处理录制的音频文件，并手动验证结果以展示其准确性。我们选择了两次总结汇报来比较不同场次之间的说话人识别结果，旨在展示互动模式的异同。
分析了10次总结汇报，每次持续约30分钟。经过数据处理，录制的数据实现了说话人识别，进而有助于自动创建可视化的互动模式，如社会关系图。研究结果和数据可视化展示了实现基于音频的互动模式可视化的技术可行性，平均准确率为97.78%。我们进一步分析了两个不同的总结汇报案例，以发现场次之间的异同。通过量化参与者的回应率，我们能够确定并量化每次总结汇报中教员引发的互动模式水平。在一场次中，总结汇报者引发了学生28%的反馈，而在另一场次中，这一比例增至36%。
我们的结果表明，说话人识别技术可以准确且自动地应用于提供总结汇报互动的可视化。这种应用对模拟教育师资的发展可能有益。这些可视化可以支持教员促进和评估总结汇报环节，最终提高基于模拟的医疗保健教育中的学习成果。
