Update app.py

app.py

@@ -67,15 +67,18 @@ 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):
+def transcribe_audio(audio_file, batch_size=16, sample_rate =16000):
     # --audio file to tensor
-    waveform, sample_rate = torchaudio.load(audio_file)
-    samples = waveform.numpy()
+    waveform = waveform.numpy(audio_file)
+    samples = samples.numpy(audio_file)
+    if waveform.ndim > 1:
+        waveform = waveform[0, :]
+    waveform, samples = torchaudio.load(waveform, samples)
 
     start_time = time.time()
 
     # --pipe it
-    outputs = pipe(samples, sampling_rate=sample_rate, batch_size=batch_size, return_timestamps=False)
+    outputs = pipe(waveform, samples, sampling_rate=sample_rate, batch_size=batch_size, return_timestamps=False)
     text = outputs["text"]
 
     end_time = time.time()