Multi-class Decoding of Attended Speaker Direction Using Electroencephalogram and Audio Spatial Spectrum