update

examples/fsmn_vad_by_webrtcvad/yaml/config-240-80-4.yaml

@@ -8,7 +8,7 @@ hop_size: 80
 win_type: hann
 
 # model
-fsmn_input_size: 257
+fsmn_input_size: 129
 fsmn_input_affine_size: 140
 fsmn_hidden_size: 250
 fsmn_basic_block_layers: 4

examples/fsmn_vad_by_webrtcvad/yaml/config-256-128-4.yaml

@@ -8,7 +8,7 @@ hop_size: 128
 win_type: hann
 
 # model
-fsmn_input_size: 257
+fsmn_input_size: 129
 fsmn_input_affine_size: 140
 fsmn_hidden_size: 250
 fsmn_basic_block_layers: 4

examples/fsmn_vad_by_webrtcvad/yaml/config-512-256-4.yaml

@@ -8,7 +8,7 @@ hop_size: 256
 win_type: hann
 
 # model
-fsmn_input_size: 513
+fsmn_input_size: 257
 fsmn_input_affine_size: 140
 fsmn_hidden_size: 250
 fsmn_basic_block_layers: 4

toolbox/torchaudio/models/vad/fsmn_vad/modeling_fsmn_vad.py

@@ -282,7 +282,14 @@ class FSMNVadModelExport(nn.Module):
 
 
 def main1():
-    config = FSMNVadConfig()
+    config = FSMNVadConfig(
+        nfft=256,
+        win_size=256,
+        hop_size=128,
+        fsmn_input_size=129,
+        fsmn_input_affine_size=140,
+        fsmn_hidden_size=250
+    )
     model = FSMNVadPretrainedModel(config=config)
 
     noisy = torch.randn(size=(1, 16000), dtype=torch.float32)
@@ -349,4 +356,4 @@ def main2():
 
 
 if __name__ == "__main__":
-    main2()
+    main1()