debug extract per speaker

nnet/xsets.py

@@ -478,6 +478,7 @@ class IdMapSetPerSpeaker(Dataset):
                  data_root_path,
                  file_extension,
                  transform_pipeline={},
+                 transform_number=1,
                  frame_rate=100,
                  min_duration=0.165
                  ):
@@ -494,7 +495,6 @@ class IdMapSetPerSpeaker(Dataset):
         self.data_root_path = data_root_path
         self.file_extension = file_extension
         self.len = len(set(self.idmap.leftids))
-        self.transformation = transform_pipeline
         self.min_duration = min_duration
         self.sample_rate = frame_rate
         self.speaker_list = list(set(self.idmap.leftids))
@@ -503,13 +503,8 @@ class IdMapSetPerSpeaker(Dataset):
         self.output_im.rightids = self.output_im.leftids
         self.output_im.start = numpy.empty(self.output_im.rightids.shape[0], "|O")
         self.output_im.stop = numpy.empty(self.output_im.rightids.shape[0], "|O")
-
-        self.transform = []
-        #if (len(self.transformation) > 0):
-        #    if (self.transformation["pipeline"] != '') and (self.transformation["pipeline"] is not None):
-        #        self.transform_list = self.transformation["pipeline"].split(',')
-        if self.transformation is not None:
-            self.transform_list = self.transformation.split(",")
+        self.transformation = transform_pipeline
+        self.transform_number = transform_number
 
         self.noise_df = None
         if "add_noise" in self.transform:
@@ -555,10 +550,10 @@ class IdMapSetPerSpeaker(Dataset):
         speech = torch.cat(tmp_data, dim=1)
         speech += 10e-6 * torch.randn(speech.shape)
 
-        if len(self.transform) > 0:
+        if len(self.transformation.keys()) > 0:
             speech = data_augmentation(speech,
                                        speech_fs,
-                                       self.transform,
+                                       self.transformation,
                                        self.transform_number,
                                        noise_df=self.noise_df,
                                        rir_df=self.rir_df)

nnet/xvector.py

@@ -1773,7 +1773,7 @@ def extract_embeddings_per_speaker(idmap_name,
                                  file_extension=file_extension,
                                  transform_pipeline=transform_pipeline,
                                  frame_rate=sample_rate,
-                                 min_duration=(model.context_size() + 2) * frame_shift * 2)
+                                 min_duration=1.)
 
     dataloader = DataLoader(dataset,
                             batch_size=1,
@@ -1788,10 +1788,7 @@ def extract_embeddings_per_speaker(idmap_name,
 
         # Get the size of embeddings to extract
         name = list(model.before_speaker_embedding.state_dict().keys())[-1].split('.')[0] + '.weight'
-        if extract_after_pooling:
-            emb_size = model.before_speaker_embedding.state_dict()[name].shape[1]
-        else:
-            emb_size = model.embedding_size
+        emb_size = model.embedding_size
 
         # Create the StatServer
         embeddings = StatServer()