use 4s samples

app.py

@@ -141,7 +141,9 @@ def model_infernce(inputs):
         resampler = T.Resample(sample_rate, resample_rate)
         waveform = resampler(waveform)
     
-    waveform = waveform.view(-1,) # make it (n_sample, )
+    #waveform = waveform.view(-1,) # make it (n_sample, )
+    waveform = waveform[0][0:4*resample_rate]  # cut to 4s samples
+    
     model_inputs = processor(waveform, sampling_rate=resample_rate, return_tensors="pt")
     model_inputs.to(device)
     with torch.no_grad():