We compared the proposed method and the other method (under three conditions) for the above three simulations. As shown by the results of Figs. 11, 14, and 18, the segregation accuracy using the proposed method was better than the other three conditions. Comparing the proposed method and condition 1 shows the advantage of the constraint of gradualness of change (smoothness). Comparing the proposed method and condition 2 shows the advantage of constraining the gradualness of change (smoothness) and harmonicity and also shows the segregation accuracy in the same frequency region in which concurrent harmonic components exist. Comparing the proposed method and condition 3 shows the improved accuracy of the proposed method. The result for these three simulations and three conditions show that the proposed method can segregate the desired signal from a mixed signal, with about 4 dB better precision and about 10 dB improvement in spectrum distortion.
As the above results, the problem of segregating the desired vowel from noisy vowel, using constraints related to Bregman's regularities is examined in this section.