Fig. 2From: Non-parallel dictionary learning for voice conversion using non-negative Tucker decompositionDecomposition of spectrograms to frequency bases and phonemic information using NTDBack to article page