In the rapidly evolving landscape of higher education and adult learning, asynchronous video-based online learning has not only become the new norm but has also emerged as the cornerstone of instructional delivery for Massive Open Online Courses (MOOCs). Despite its widespread adoption, this learning mode confronts a critical challenge: the inherent lack of social presence, posing a significant risk of diminishing learner affective engagement and, consequently, jeopardizing the efficacy of learning outcomes. Addressing this pressing issue, our study conducted a comprehensive analysis of 240 instructional videos from 240 distinct instructors on a MOOC platform, supplemented by 845 post-course learner feedback surveys from a diverse cohort of college students and adult learners. Using deep learning and statistical analysis, the research revealed that the on-screen presence of instructors does not inherently affect students' affective engagement. The study revealed that learners' affective engagement is affected by distinct combinations of the instructor's facial and paraverbal expressions, including happiness, surprise, and anger, which vary depending on whether the instructor is visible. The discovery that vocal attractiveness is a pivotal element in enhancing learners' affective engagement with instructional videos marks a paradigm shift in our understanding of digital andragogy and heutagogy. This study propels academic discourse by illuminating the critical role of instructor non-verbal cues in establishing social presence and facilitating emotional contagion within asynchronous video-based online learning but also provides educators and content creators with empirically-backed techniques to revolutionize video instruction and amplify affective engagement.