Fixed app

app.py

@@ -42,7 +42,8 @@ def waveformer(audio, label_choices):
     fs, mixture = audio
     if fs != 44100:
         raise ValueError(fs)
-    mixture = torch.from_numpy(mixture).unsqueeze(0)
+    mixture = torch.from_numpy(
+        mixture).unsqueeze(0).unsqueeze(0).to(torch.float)
 
     # Construct the query vector
     if len(label_choices) == 0:
@@ -54,7 +55,7 @@ def waveformer(audio, label_choices):
     with torch.no_grad():
         output = model(mixture, query)
 
-    return fs, output.squeeze(0).numpy()
+    return fs, output.squeeze(0).squeeze(0).to(torch.short).numpy()
 
 
 label_checkbox = gr.CheckboxGroup(choices=TARGETS)

requirements.txt

@@ -6,4 +6,5 @@ soundfile
 numpy
 speechbrain
 wget
+torchmetrics