Overview: Speech Enhancement with GANs

In this chapter, we will learn how to implement a speech enhancement generative adversarial network (SEGAN), a framework for audio denoising and speech enhancement using GANs, developed by Santiago Pascual and others. We will learn to train the model on multiple speakers and noise conditions. We will learn to evaluate the model qualitatively and quantitatively. By the end of the chapter, we will have learned to train and have a model for speech enhancement.

At the end of this chapter, we will be able to implement our own speech enhancement model and evaluate its performance.

Topics covered in this chapter

The following topics will be covered in this chapter:

• Audio enhancement

• Experimental setup

• Model implementation: SEGAN

• Qualitative and quantitative evaluation