merge

s4d/clustering/hac_iv.py

@@ -41,7 +41,6 @@ def hac_iv(diar, scores, method="complete", threshold=0.0):
     lscores = copy.deepcopy(scores)
     # get the triangular part of the distances
     distances, t = scores2distance(lscores, threshold)
-
     # distance = numpy.copy((scores.scoremat + scores.scoremat.T) / 2.0) * -1.0
     # numpy.fill_diagonal(distance, numpy.inf)
     # min = numpy.min(distance)

s4d/diar.py

@@ -984,7 +984,7 @@ class Diar():
             channel = 'U'
             if diar._attributes.exist('channel'):
                 channel = segment['channel']
-            if time_float:
+            if not time_float:
                 lst.append('{:s} 1 {:d} {:d} {:s} {:s} {:s} {:s}\n'.format(
                     segment['show'], segment['start'], segment['stop'] - segment['start'], gender,
                     channel, env, segment['cluster']))

s4d/nnet/seqtoseq.py

@@ -61,7 +61,7 @@ def save_checkpoint(state, is_best, filename='checkpoint.pth.tar', best_filename
         shutil.copyfile(filename, best_filename)
 
 class PreNet(nn.Module):
-    def __init(self,
+    def __init__(self,
                sample_rate=16000,
                windows_duration=0.2,
                frame_shift=0.01):
@@ -153,16 +153,11 @@ class SeqToSeq(nn.Module):
                  linear_1,
                  linear_2,
                  output_size=1):
-        """
 
-        :param input_size:
-        :param lstm_1:
-        :param lstm_2:
-        :param linear_1:
-        :param linear_2:
-        :param output_size:
-        """
-        super(BLSTM, self).__init__()
+        super(SeqToSeq, self).__init__()
+        self.preprocessor = PreNet(sample_rate=16000,
+                                   windows_duration=0.2,
+                                   frame_shift=0.01)
 
         self.lstm_1 = nn.LSTM(input_size, lstm_1 // 2, bidirectional=True, batch_first=True)
         self.lstm_2 = nn.LSTM(lstm_1, lstm_2 // 2, bidirectional=True, batch_first=True)