A Real Time Implementation of SpectroNet for Music Player Application

Keyword Spotting(KWS), which refers to the task of recognizing spoken words out of a small database of words, has become one of the most popular research topics. After the emergence of deep learning architectures, such as Convolutional Neural Network(CNN) and Reccurent Neural Network(RNN), works have been done to apply these architectures to Keyword Spotting problems. Keyword Spotting can have different topics/themes, such as home automation, robot control, etc. In this paper, music player is chosen as the topic of the KWS. A novel database will be created to test the system. The network chosen to implement this is SpectroNet, which is a low-complexity hybrid CNN-RNN architecture based on the DenseNet architecture.