recognizer.py 文件源码

def get_mfcc_feat(self):
        # creating codebook with all models

        mfcc_feats = None

        for filename in glob.iglob('../data/voices/*.wav'):
            print filename
            (rate, sig) = wav.read(filename)

            # MFCC Features. Each row corresponds to MFCC for a frame
            mfcc_person = mfcc(sig.astype(np.float64), rate)

            if mfcc_feats is None:
                mfcc_feats = mfcc_person 
            else:
                mfcc_feats = np.concatenate((mfcc_feats, mfcc_person), axis=0)

        # Normalize the features
        whitened = whiten(mfcc_feats)
        self.codebook, labeled_obs = kmeans2(data=whitened, k=3)