Avaliação de técnicas espectrais aplicadas à remoção de ruído em sinais de áudio musical

• Autor(a) principal: Bianca, Flávio Giraldeli
• Data de Publicação: 2009
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Federal do Espírito Santo (riUfes)
• Texto Completo: http://repositorio.ufes.br/handle/10/4076
• topic: Ruído musical; Subtração espectral; Análise espectral; Wavelets (Matemática); Processamento de sinais - Técnicas digitais; Som - Registro e reprodução - Técnicas digitais; Controle de ruído; Medidas Elétricas, Magnéticas e Eletrônicas; Instrumentação; 621.3

This work evaluated the problem of noise removal in one-dimensional signals, especially music signals. Three techniques with good results reported in the literature were selected: Wavelet Thresholding, Time-Frequency Block Thresholding, both based on the multiresolution theory, in addition to the classical Spectral Subtraction technique, widely used in speech signals, which was here modified to take into account musical audio signals. In addition to the capability to remove noise, each technique was also evaluated by their capability to not introduce artifacts in the filtered signal like the musical noise, which is characterized by “musical notes” that does not exist in the original audio. In order to evaluate the results, a set of signals was selected for the tests. First, each technique was tested with synthetic signals that do not characterize musical audio. Then, the techniques were evaluated for classical and popular music. For each case, the results are extensively discussed. In each case, it was measured the Signal to Noise Ratio (SNR) in order to compare the performance of the techniques. Finally, subjective tests were implemented, by interviews with 10 volunteers, in order to evaluate the subjective quality of the results of each technique. The results showed that techniques that have yielded good results for synthetic signals, not necessarily are suitable to the musical quality demanded by listeners. Also, the modification proposed in the Spectral Subtraction technique was enough to place it, in a subjective quality rank, between Time-Frequency Block Thresholding (best results) and Wavelet Thresholding (worst results).