GRaNN: feature selection with golden ratio-aided neural network for emotion, gender and speaker identification from voice signals