diff --git a/lerobot/common/robot_devices/control_utils.py b/lerobot/common/robot_devices/control_utils.py
index e49a4e71..111e21f0 100644
--- a/lerobot/common/robot_devices/control_utils.py
+++ b/lerobot/common/robot_devices/control_utils.py
@@ -117,8 +117,9 @@ def predict_action(observation, policy, device, use_amp):
             if "image" in name:
                 observation[name] = observation[name].type(torch.float32) / 255
                 observation[name] = observation[name].permute(2, 0, 1).contiguous()
-            # Convert to pytorch format: channel first and float32 in [-1,1] (always the case here) with batch dimension
+            # Convert to pytorch format: channel first and float32 in [-1,1] with batch dimension
             if "audio" in name:
+                observation[name] = observation[name].type(torch.float32)
                 observation[name] = observation[name].permute(1, 0).contiguous()
             observation[name] = observation[name].unsqueeze(0)
             observation[name] = observation[name].to(device)