diff --git a/enhancer/models/demucs.py b/enhancer/models/demucs.py
index bf9d429..5d7e99f 100644
--- a/enhancer/models/demucs.py
+++ b/enhancer/models/demucs.py
@@ -49,7 +49,7 @@ class DemucsEncoder(nn.Module):
         self.encoder = nn.Sequential(
             nn.Conv1d(num_channels, hidden_size, kernel_size, stride),
             nn.ReLU(),
-            nn.Conv1d(hidden_size, hidden_size * multi_factor, kernel_size, 1),
+            nn.Conv1d(hidden_size, hidden_size * multi_factor, 1, 1),
             activation,
         )
 
@@ -72,7 +72,7 @@ class DemucsDecoder(nn.Module):
         activation = nn.GLU(1) if glu else nn.ReLU()
         multi_factor = 2 if glu else 1
         self.decoder = nn.Sequential(
-            nn.Conv1d(hidden_size, hidden_size * multi_factor, kernel_size, 1),
+            nn.Conv1d(hidden_size, hidden_size * multi_factor, 1, 1),
             activation,
             nn.ConvTranspose1d(hidden_size, num_channels, kernel_size, stride),
         )
@@ -116,7 +116,7 @@ class Demucs(Model):
     ED_DEFAULTS = {
         "initial_output_channels": 48,
         "kernel_size": 8,
-        "stride": 1,
+        "stride": 4,
         "depth": 5,
         "glu": True,
         "growth_factor": 2,
@@ -179,7 +179,7 @@ class Demucs(Model):
                 num_channels=num_channels,
                 hidden_size=hidden,
                 kernel_size=encoder_decoder["kernel_size"],
-                stride=1,
+                stride=encoder_decoder["stride"],
                 glu=encoder_decoder["glu"],
                 layer=layer,
             )
@@ -236,7 +236,7 @@ class Demucs(Model):
                 self.hparams.sampling_rate,
             )
 
-        return x
+        return x[..., :length]
 
     def get_padding_length(self, input_length):