Uso de engenharia de características para predição da duração do compasso musical do forró

Abstract

For humans, dancing is highly associated with musicality and the rhythm of songs, which implies that dance participants need to hear the music in order to interact with each other. However, deaf or hearing impaired people (D/HH) are often excluded from this type of social interaction environment, making it even more difficult for them to live in society. Thus, it is necessary to use techniques such as an ANN to convert sound signals into others that D/HH can perceive. This technique already exists in the literature, using the Fourier transform in part of the frequency spectrum of the music as input of an ANN to determine the base step tempo of forró music, however, it was not analyzed how the variations in the network input affect its performance. Therefore, this work proposes the variation of the number of inputs and the frequency spectrum to determine the best extraction of music characteristics. With this, an extraction of audio features was defined that reduced the error by more than 20% compared to the error of the technique existing in the literature that can be used for application in a tool capable of inserting the deaf and hearing impaired in the forró dance environment, providing a better coexistence of these individuals in the local society.