update

egs/aishell/paraformer/run.sh

@@ -3,8 +3,8 @@
 . ./path.sh || exit 1;
 
 # machines configuration
-CUDA_VISIBLE_DEVICES="0"
-gpu_num=1
+CUDA_VISIBLE_DEVICES="0,1"
+gpu_num=2
 count=1
 gpu_inference=true  # Whether to perform gpu decoding, set false for cpu decoding
 # for gpu decoding, inference_nj=ngpu*njob; for cpu decoding, inference_nj=njob

funasr/datasets/small_datasets/dataset.py

@@ -111,6 +111,7 @@ class ESPnetDataset(Dataset):
             int_dtype: str = "long",
             dest_sample_rate: int = 16000,
             speed_perturb: tuple = None,
+            mode: str = "train",
     ):
         assert check_argument_types()
         if len(path_name_type_list) == 0:
@@ -125,6 +126,7 @@ class ESPnetDataset(Dataset):
         self.int_dtype = int_dtype
         self.dest_sample_rate = dest_sample_rate
         self.speed_perturb = speed_perturb
+        self.mode = mode
 
         self.loader_dict = {}
         self.debug_info = {}
@@ -148,7 +150,8 @@ class ESPnetDataset(Dataset):
             loader_type:  loader_type. sound, npy, text, etc
         """
         if loader_type == "sound":
-            loader = SoundScpReader(path, self.dest_sample_rate, normalize=True, always_2d=False, speed_perturb=self.speed_perturb)
+            speed_perturb = self.speed_perturb if self.mode == "train" else None
+            loader = SoundScpReader(path, self.dest_sample_rate, normalize=True, always_2d=False, speed_perturb=speed_perturb)
             return AdapterForSoundScpReader(loader, self.float_dtype)
         elif loader_type == "kaldi_ark":
             loader = kaldiio.load_scp(path)