Separación ciega de fuentes no-determinada aplicada a mezclas de voz con base en la transformada wavelet discreta

blind source separation, bss, is a signal processing technique which estimates sources from linearly mixed signals and it uses methods such as ica for sources that are statistically independent. among the best known bss algorithms is the jade method, which requires that the number of independent signals match the number of observed signals (sensors). in the real world, the number of sensors is lower than the number of sources (undetermined bss) and therefore the problem has no solution. this work proposes a solution for undetermined bss by pre-processing and decomposition stages based on the discrete wavelet transform (dwt). our proposal, which it is known as dwt+bss, creates a virtual observed signal from a real observed signal and it uses the wavelet coefficients of the observed signals as the inputs of the classical jade algorithm. we validated our model with voice and audio signals obtaining indexes of similarity over 0.7 between the original and the estimated sources.