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