Figure 9.

The proposed Fast Neural Style Transfer architecture. Based on sample data from NinaPro database and a PD’s patient EMG data, we train a transformer network that is capable of applying the tremor pattern on any input EMG signal. The discriminator trained during the DCGAN steps was used as a feature extractor for the Style Loss calculation, while the Content Loss was calculated based on MSE from the content signal plus a custom-designed loss for applying penalties where the EMG content signal has higher amplitudes.