Test removing waveform_tensor and using data as input

app.py

@@ -200,13 +200,14 @@ def align_timestamps(segments, dia_seg_last_end):
 
 def transcribe_diarize(audio):
     sr, data = audio
-    processed_data = np.array(data).astype(np.float32) / 32767.0
-    waveform_tensor = torch.tensor(processed_data[np.newaxis, :])
+    #processed_data = np.array(data).astype(np.float32) / 32767.0
+    #waveform_tensor = torch.tensor(processed_data[np.newaxis, :])
 
     transcription_res = transcribe(sr, data)
     # results from the diarization pipeline
     diarization_res = diarization_pipeline(
-        {"waveform": waveform_tensor, "sample_rate": sr}
+       #Test directly with processed_data {"waveform": waveform_tensor, "sample_rate": sr}
+        {"waveform": data, "sample_rate": sr}
     )
     dia_seg, dia_label = [], []
     for segment, _, label in diarization_res.itertracks(yield_label=True):