This chapter focuses on the semantic-aware source compression coding methods, covering AI-driven encoding techniques for speech, image, and video modalities in greater depth. For AI speech codecs, this chapter primarily introduces the VQ-VAE-based end-to-end AI speech codec, the LPC-enhanced AI speech codec, and the text-based AI speech codec, while also explaining their respective strengths, design motivations, and potential application scenarios. For AI image codecs, this chapter highlights how data-driven designs in transformation coding and entropy modeling replace handcrafted components to enable end-to-end rate-distortion optimization, and it further discusses how innovations such as hyperpriors, context models, and Transformer-based architectures capture both local and global dependencies, thereby improving compression efficiency, perceptual fidelity, adaptability, and decoding speed. For AI video codecs, this chapter explains the process of traditional video compression standards and the metrics of video quality, and based on these backgrounds, it introduces the architecture, design principles, and emerging trends of the existing AI video codec.