What kind of features do you choose when doing audio recognition in Caffe, MFCC,wavelet,or other?

[Fei Kong]

I wanna do audio recognition with the dataset of general artificial sound ，and I realize that the sound should be represented by some kind of features(for example,MFCC ,wavalet ,spectrogram and so on) extracted from it, but I don't know what kind of features I should choose.I wanna know what is the common practice in this field.