Text this: Multi-Class Decoding of Attended Speaker Direction Using Electroencephalogram and Audio Spatial Spectrum