Classificação de segmentos de vídeos por trechos de fala utilizando rede neural convolucional

Abstract

Videos are one of the most popular contents on the internet, social medias platforms and cell phone’s made capture and publish videos so easy and fast. However, in a lot of cases it’s necessary edit the content manually, which is irksome, because it takes a lot of time. Therefore, the need automatic editing alternative. In this final paper presents a creation proposal of video classification model, using convolutional neural networks. Starts with several self-authored videos, which each one was randomly cut in different parts. From these parts, there were took the audios in wav (waveform audio format), previously classified and after extracted the characteristics of each file using the Mel frequency cepstrum coefficients method (MFCC), it was got spectrograms as like 2D images. Those spectrograms start the convolutional neural networks implemented producing as output, a classification model. Experimental results from the methodology proposed, showed that this model it’s able to classify 92,52% of videos stretches correctly.