Update app.py

app.py

@@ -70,26 +70,24 @@ 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):
-
-    with tempfile.TemporaryDirectory() as tmpdirname:
-        temp_path = Path(tmpdirname) / "uploaded_audio.wav"
-        
-        with open(temp_path, "wb") as f:
-            f.write(audio_file.read()) 
-        
-    waveform, sample_rate = torchaudio.load(str(temp_path))
-        
-    samples = waveform.numpy()
-
+def transcribe_audio(audio_file, filepath, batch_size=16, sample_rate=16000):
+    
+    #waveform, sample_rate = torchaudio.load(str(audio_file) #3
+    audio_file = filepath(audio_file)
+    waveform, sample_rate = torchaudio.load(audio_file)  #1
+    #waveform, sample_rate = torchaudio.load("{filepath}") #2
+           
+    # --convert to mono
+    if waveform.ndim > 1:
+        waveform = waveform[0, :]
+
+    waveform = waveform.numpy()
+
+    start_time = time.time()
     
-    if samples.ndim > 1:
-        samples = samples[0, :]
-        
-
     # --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()