Update app.py

app.py

@@ -69,18 +69,18 @@ pipe = pipeline("automatic-speech-recognition", model="NbAiLab/nb-whisper-large"
 @spaces.GPU()
 def transcribe_audio(audio_file, batch_size=16, sample_rate =16000):
     # --audio file to tensor
-    
+    waveform = waveform.numpy(audio_file)
+    samples = samples.numpy(audio_file)
     waveform, samples = torchaudio.load(waveform, samples)
+    
     if waveform.ndim > 1:
         waveform = waveform[0, :]
-    waveform = waveform.numpy(audio_file)
-    samples = samples.numpy(audio_file)
 
     start_time = time.time()
 
     # --pipe it
-    outputs = pipe(waveform, samples, sampling_rate=sample_rate, batch_size=batch_size, return_timestamps=False)
-    text = outputs["text"]
+    with torch.no_grad():
+        outputs = pipe(waveform, samples, sampling_rate=sample_rate, batch_size=batch_size, return_timestamps=False)
 
     end_time = time.time()