hac.HACFeatureExtractor( filenames, ks, lags=[5,2], name="HACFeatureExtracter", **mfcc_params )
hac.HACFeatureExtractor is a module for acoustic feature extraction based on the histogram of acoustic co-occurrences (HAC).
It extracts HAC features from wave-format files.
See here for details about HAC.
| Parameter | Type | Description |
|---|---|---|
| filenames | array | Paths to wave files |
| ks | array | Number of elements in each code book (triplet of int) |
| lags | array | List of lags to compute co-occurrence of acoustic events (the corresponding histograms are concatenated) |
| name | string | Module name |
| mfcc_params | tuple | Parameters for computing mfcc features (used librosa) default “n_mfcc”: 13 “n_fft”: 2048 “hop_length”: 512 “n_mels”: 128 |
# import necessary modules
import hac
import mlda
# make a list of paths to wav data
wavs = ["./data00.wav", "./data01.wav", "./data02.wav",
"./data03.wav", "./data04.wav", "./data05.wav",
"./data06.wav", "./data07.wav", "./data08.wav"]
# define the modules
obs = hac.HACFeatureExtractor( wavs, [10,10,10], lags=[5] ) # convert wav data into hac features
mlda1 = mlda.MLDA( 3, [100], category=[0,0,0,1,1,1,2,2,2] ) # classify into three classes
# construct the model
mlda1.connect( obs ) # connect obs to mlda1
mlda1.update() # train mlda1