Update app.py

app.py

@@ -16,7 +16,7 @@ SR = 16000
 
 
 def resample(speech_array, sampling_rate):
-    speech = torch.from_numpy(np.array([speech_array])).float()
+    speech = torch.from_numpy(speech_array)
     print(speech, speech.shape, sampling_rate)
     resampler = torchaudio.transforms.Resample(sampling_rate)
     speech = resampler(speech).squeeze().numpy()
@@ -47,6 +47,7 @@ print(device)
 
 def recognize(audio):
     sr, audio_array = audio
+    audio_array = audio_array.astype(np.float32)
     print(sr, audio_array)
     return predict(audio_array, sr)