自己報告と生体信号を活用したストレスマネジメント介入における助言選好への性格特性の影響：チャットボット設計に向けた検討

抄録

Problems such as the difficulty of performance and time and space constraints have existed in conventionaltherapists interview. In this study, we developed a spoken dialogue system to support therapists interview practiceand evaluated its effectiveness. This system aims to overcome these problems and provide an environment wheretherapists can practise anytime, anywhere. The system is built on Remdis, a platform for developing text, speech andmultimodal dialogue systems. Specifically, the system is implemented by using a large scale language model (LLM)to simulate the patient role and combining techniques such as speech recognition (ASR), turn-taking and speech synthesis(TTS). As a result of the evaluation experiment with university students and therapists, this system was highlyevaluated in the following points. At first, it was confirmed that the intuitive operability of the system with speechinput, the quality of the speech synthesis, and the clear and easy-to-understand responses by ChatGPT contributedto the facilitated dialogue. In addition, the experimental participants evaluated that this system is sufficiently usefulas a tool for interview practice in the medical field and that it has high applicability to actual practice. Furthermore,the speech recognition rate and the success rate of response generation showed a high level, and the reliability of thebasic performance was established. The proposed system shows the potential to be an effective tool in the educationof therapists.