classification - How to use MFCC vectors for classifying a single audio file? -
this silly question, couldn't find details anywhere.
so have audio recording (wav file) 3 seconds long. sample , needs classified [class_a] or [class_b].
by following tutroial on mfcc, divided sample frames (291 frames exact) , i've gotten mfccs each frame.
now have 291 feature vectors, length of each vector 13.
my question is; how use vectors classifier (k-nn example)? have 291 vectors represent 1 sample. know how work 1 vector 1 sample, don't know if have 291 of them. couldn't find explanation anywhere.
each of vectors represent spectral characteristics of audio file, varies in time. depending on length of frames, might want group of them (for example averaging dimension) match resolution want classifier work. example, think of particular sound might have envelope attack time of 2ms: may fine-grained want time quantization a) group , average number of mfcc vectors represent 2ms; or b) recompute mfccs desired time resolution.
if want keep resolution fine, can concatenate 291 vectors , treat single vector (of 291 x 13 dimensions), need huge dataset train on.
Comments
Post a Comment