def getDist(args):
sample = args['s_path']
word = args['w_path']
model = args['model']
trans = args['trans']
sample_wav, fs = walkman.load(sample)
sample_feat = logfbank(sample_wav,
samplerate=fs,
lowfreq=64,
highfreq=8000,
nfilt=40,
nfft=2048)
sample_feat_trans = featureTransform(sample_feat, trans)
word_wav, fs = walkman.load(word)
word_feat = logfbank(word_wav,
samplerate=fs,
lowfreq=64,
highfreq=8000,
nfilt=40,
nfft=2048)
word_feat_trans = featureTransform(word_feat, trans)
y_sample = model.predict(sample_feat_trans)
y_word = model.predict(word_feat_trans)
dist, _, _, _ = dtw(makeList(y_word), makeList(y_sample), euclidean)
args['dist'] = dist
return args
def score(sample, truth, model, trans):
sample_wav, fs = walkman.load(sample)
sample_feat = logfbank(sample_wav, samplerate=fs, lowfreq=64, highfreq=8000, nfilt=40)
sample_feat_trans = featureTransform(sample_feat, trans)
print(np.shape(sample_feat_trans))
x = sample_feat_trans
y_sample = model.predict(x, batch_size=512)
y_truth = load_truth(truth)
dist, _, _, _ = dtw(makeList(y_truth), makeList(y_sample), klDist)
return dist
def getDtw(args):
dist, _, _, _ = dtw(makeList(args['w_feat']), makeList(args['s_feat']), euclidean)
args['dist'] = dist
return args