predict smoothing

[asr1617data.git] / predict.py

diff --git a/predict.py b/predict.py
index b561c65..3e0e2f7 100644 (file)
--- a/predict.py
+++ b/predict.py
@@ -1,6 +1,9 @@
-import numpy as np
 import sys
 
+if len(sys.argv) != 3:
+    print("Usage: {} MODEL.json AUDIO.wav".format(sys.argv[0]))
+    sys.exit(1)
+
 import pympi
 import scipy.io.wavfile as wav
 import numpy as np
@@ -16,10 +19,12 @@ with open(modelfile, 'r', encoding='utf-8') as f:
 model = model_from_json(json)
 model.load_weights(hdf5file)
 
-(_, winlen, winstep, _) = modelfile.split('_')
+(_, winlen, winstep, name) = modelfile.split('_')
 winlen = float(winlen)
 winstep = float(winstep)
 
+multiclass = name == 'multi'
+
 model.compile(
     loss='binary_crossentropy',
     optimizer='rmsprop',
@@ -30,16 +35,20 @@ data = mfcc(sig, rate, winlen, winstep, numcep=13, appendEnergy=True)
 tgob = pympi.TextGrid(xmax=winstep*len(data))
 tier = tgob.add_tier('lyrics')
 
-time = 0.0
-lastlabel = False
-lasttime = 0.0
-for i in model.predict(data, batch_size=32, verbose=0):
-#    print('{}\t{}'.format(time, i))
-    label = i > 0.5
-    if label != lastlabel and time-lasttime > 0.5:
-        tier.add_interval(lasttime, time, '*' if lastlabel else '')
-        lastlabel = label
-        lasttime = time
-
-    time += winstep
+window_len = int(1.0/winstep)
+
+x = model.predict(data, batch_size=32, verbose=0)
+s = np.r_[x[window_len-1:0:-1],x,x[-2:-window_len-1:-1]]
+w = np.hanning(window_len)
+#
+smoothed = np.convolve(w/w.sum(), s[:,0], mode='valid')
+wavdata = np.uint8(list(map(int,
+    smoothed*255))[int(window_len/2):-1*(int(window_len/2))])
+#wavdata = np.uint8(x*255)
+
+print('sr: ', int(1.0/winstep))
+print("len(wavdata): ", len(wavdata))
+print("len(x): ", len(x))
+wav.write('class.wav', int(1.0/winstep), wavdata)
+
 tgob.to_file('/dev/stdout')