Articulatory Controllable Speech Modification based on Gaussian Mixture Models with Direct Waveform Modification using Spectrum Differential