merge and sidesampler

nnet/preprocessor.py

@@ -211,6 +211,7 @@ class MelSpecFrontEnd(torch.nn.Module):
         """
 
         :param x:
+        :param is_eval:
         :return:
         """
         with torch.no_grad():

nnet/xsets.py

@@ -154,7 +154,7 @@ class SideSampler(torch.utils.data.Sampler):
         return iter(self.index_iterator)
 
     def __len__(self) -> int:
-        return (self.samples_per_speaker * self.spk_count * self.examples_per_speaker) // self.num_replicas
+        return (self.samples_per_speaker * self.spk_count * self.examples_per_speaker * self.num_replicas) // self.num_process
 
     def set_epoch(self, epoch: int) -> None:
         self.epoch = epoch

nnet/xvector.py

@@ -863,9 +863,9 @@ def update_training_dictionary(dataset_description,
     dataset_opts["batch_size"] = 64
 
     dataset_opts["train"] = dict()
-    dataset_opts["train"]["duration"] = 2.
+    dataset_opts["train"]["duration"] = 4.
     dataset_opts["train"]["chunk_per_segment"] = -1
-    dataset_opts["train"]["overlap"] = 1.9
+    dataset_opts["train"]["overlap"] = 3.9
     dataset_opts["train"]["sampler"] = dict()
     dataset_opts["train"]["sampler"]["examples_per_speaker"] = 1
     dataset_opts["train"]["sampler"]["samples_per_speaker"] = 100
@@ -944,7 +944,6 @@ def update_training_dictionary(dataset_description,
     training_opts["best_model_name"] = "best_model.pt"
     training_opts["checkpoint_frequency"] = "10"
 
-
     # Use options from the YAML config files
     fill_dict(dataset_opts, tmp_data_dict)
     fill_dict(model_opts, tmp_model_dict)