Edge Impulse-based automated machine learning for comedy performance evaluation

Abstract

The evaluation of comedic performances presents a challenge due to the subjectivity of humor and various performance factors. This paper proposes a potential solution to this problem by utilizing artificial intelligence (AI), particularly automated machine learning (AutoML), to develop an automatic comedy show scoring system using a diverse Chinese acoustic dataset and the Edge Impulse platform. The proposed method involves analyzing complete video recordings of performances by the renowned Chinese comedy troupe, Deyunshe, to assess audience responses and critical acoustic elements that contribute to a comedic performance's success. The study collected five comedic clips, totaling approximately 50 minutes, segmented and labeled them for training purposes. This study systematically explored various configurations using Edge Impulse to optimize accuracy and minimize loss, yielding encouraging results that demonstrate the potential of AI-driven scoring systems. The study identifies several avenues for future research, including enhancing the quality and quantity of training data, refining classification algorithms, and exploring alternative machine learning techniques to further improve accuracy and loss rates. Moreover, the research highlights the potential benefits of integrating real-time scoring systems, which could facilitate audience engagement and refining comedic material. This innovative approach demonstrates the potential to develop a more sophisticated, accurate, and reliable method of evaluating comedic performances, offering valuable insights into audience preferences, enabling the production of high-quality comedic content, and fostering a more objective evaluation of talent in the competitive world of comedy, which could significantly benefit the entertainment industry.