Figure 3From: Voice conversion using speaker-dependent conditional restricted Boltzmann machine Changing hidden units. The values show averaged Mel-cepstral distortion with varying numbers of hidden units J for all network-based methods (N=20,000).Back to article page