Speech synthesis, also known as text-to-speech (TTS), is the AI workload that converts written text into spoken words. In this case, the task is to generate narration audio from provided scripts for silent instructional videos.
Speech recognition performs the opposite function - it converts speech into text. Language modeling is for text understanding and prediction (e.g., GPT). Translation converts text between languages, not from text to audio.
Therefore, the most appropriate workload, according to Microsoft's AI-900 study material under the "Speech AI capabilities" section, is speech synthesis, which enables natural voice narration generation.