Robust speaker verification using short-time frequency with long-time window and fusion of multi-resolutions.