Update app.py

app.py

@@ -69,27 +69,23 @@ def convert_to_wav(filepath):
 pipe = pipeline("automatic-speech-recognition", model="NbAiLab/nb-whisper-large", chunk_length_s=30, generate_kwargs={'task': 'transcribe', 'language': 'no'})
 
 @spaces.GPU()
-def transcribe_audio(audio_file, batch_size=16, sample_rate=16000): 
-    audio_file = filepath.read(audio_file) 
-    waveform, sample_rate = torchaudio.load(audio_file) 
-    
-    if samples.ndim > 1:
-        samples = samples[0, :]
+def transcribe_audio(filepath, batch_size=16, sample_rate=16000): 
+     
+    waveform, sample_rate = torchaudio.load(filepath) 
+
+    # --convert to mono
+    if waveform.ndim > 1:
+        waveform = waveform[0, :]
 
     # --waveform to ndnumpy array
     samples = waveform.numpy()
     
     start_time = time.time()
-        
-    # --convert to mono
-    if len(samples.shape) > 1:
-        samples = samples[:, 0]
 
-    start_time = time.time()
 
     # --pipe it
     with torch.no_grad():
-        outputs = pipe(samples, sampling_rate=sample_rate, batch_size=batch_size, return_timestamps=False)
+        outputs = pipe(waveform, sampling_rate=sample_rate, batch_size=batch_size, return_timestamps=False)
 
     end_time = time.time()