Fig. 4From: Non-parallel dictionary learning for voice conversion using non-negative Tucker decompositionAverage MCD [dB] of the conventional ARBM-based VC when varying the number of hidden unitsBack to article page