Estimação de F0 via MDC aproximado e agrupamento por K-médias ponderado

Given a set of spectrum partials obtained from shorter windowed signal frames, it is proposed a fundamental frequency estimation by calculating the approximate GCD of such partials and applying a weighted K-means cluster method to classify them. Such an algorithm allows the use of shorter frame sizes, resulting in high temporal resolution for the estimation of F0 curve. The strategy produced good results when the SNR was very low (−5dB) or when the formant centers are closer to the f0 or to its first harmonics