Synthesizing photograph via Voice commands using Generative Adversarial Networks (GANs)

Abstract

Interior designers often get troubled with imagining the designs. Generative Adver- sarial Networks (GANs) can help designers put their thoughts on computer screen in

real-time by giving voice commands. GANs are one of the trending research topics in the field of artificial intelligence. Speech recognition is an important ascpet of AI in present days and GANs have the ability to generate new data based on it’s learning from gaussian curve. Synthesizing photo-realistic images is a challenging task. In this paper, An approach of synthesizing photo-realistic images from voice commands is shown. Two GAN models are used in order to generate a healthy looking image based on the voice

commands are given. Google voice API is used in order to achieve voice-to-text conver- sion. Converted text being the input for first GAN and it will generate a low-resolution

image with primitive shape conditioned with the text given. The image generated from first GAN will work as input for the second GAN along with the same text used earlier.

Second GAN will refine the image and put more details in the image along with convert- ing the image to a larger resolution. Dataset used for this purpose is created on own from

scretch, It consists of sofaset images for interior designing.