A company wants to transcribe recorded sales calls and detect the sentiment of each one. It has no machine learning engineers and no labelled training data. Which approach is most consistent with the guidance the exam expects?

• ACollect and label six months of calls, then train a custom speech and sentiment model in Amazon SageMaker.
• BPre train a foundation model on the company's call recordings so it learns the domain.
• CUse managed services, Amazon Transcribe for speech to text and Amazon Comprehend for sentiment, before considering anything custom.
• DFine tune an open source model on unlabelled recordings until accuracy is acceptable.