Update pipeline/asr_diarization.py

pipeline/asr_diarization.py

@@ -7,7 +7,6 @@ _metadata = None
 def load_asr(device="cpu", model_size="medium"):
     global _asr_model
     if _asr_model is None:
-        # force float32 so it doesn’t try to use unsupported float16 on CPU / limited GPU
         _asr_model = whisperx.load_model(model_size, device=device, compute_type="float32")
     return _asr_model
 
@@ -20,7 +19,8 @@ def transcribe_and_align(audio_path: str, device="cpu"):
         _alignment_model, _metadata = whisperx.load_align_model(
             language_code=result["language"], device=device
         )
+    # Pass device here as required
     result_aligned = whisperx.align(
-        result["segments"], _alignment_model, _metadata, audio_path
+        result["segments"], _alignment_model, _metadata, audio_path, device=device
     )
     return result, result_aligned