Singular value decomposition based low-footprint speaker adaptation and personalization for deep neural network | Synapse