In recent years, generative adversarial networks (GANs) have been increasingly applied in the field of speech signal processing. In terms of speech enhancement, GANs effectively remove noise and improve speech clarity through an adversarial training mechanism. In the field of speech conversion, GANs are capable of implementing functions such as speech emotion conversion and speech style transfer. In the field of speech synthesis, GANs can generate high-fidelity, natural and smooth speech, significantly improving the quality and diversity of speech synthesis. This chapter selects three models, namely SEGAN, CycleGAN-VC, and WaveGAN, for detailed structural and code explanations.