Analysis of Radio Speech Using Deep Neural Network

Abstract

Radio broadcasting, which uses audio information only, is still popular in the modern society in which television and movie websites are widely enjoyed. In radio programs, we generally broadcast speech sounds spoken by so called personalities. In this research, we estimated the nonlinguistic information from using machine learning, aiming at creating a audio-based broadcast program retrieving system. More concretely, it was estimated by machine learning whether the speech is a monologue that a speaker is talking alone or dialogue talked by two or more people, using a the acoustic features obtained from the speech waveforms and deep neural network.