Information fusion based on kernel entropy component analysis in discriminative canonical correlation space with application to audio emotion recognition.