Generative Adversarial Networks for Audio and Voice Enhancement
In a whole universe of multimedia applications, the observation of information tends to be a composition typically resulting from some mixing process. For some applications, such Status quo may be exactly what one wishes, however, as is not always the case, it is desirable for each generating source to be represented in a different information channel; the task carrying out such a process is known as a source separation or signal separation, regardless of what these signals represent. Such a task is not limited to a single method; there are many different ways to carry out the separation process.
In this work, GAN (Generative Adversarial Networks) frameworks were explored, coupled with more traditional techniques to perform the task of source separation aiming to provoking an audio enhancement effect for human voice signals, which includes a noise attenuation process.