add debuging for npy files

YuE/inference/infer.py

@@ -420,6 +420,12 @@ os.makedirs(recons_mix_dir, exist_ok=True)
 tracks = []
 for npy in stage2_result:
     codec_result = np.load(npy)
+    print(f"Processing {npy}")
+    print(f"  Shape: {npy_data.shape}")
+    print(f"  Min/Max: {npy_data.min()}/{npy_data.max()}")
+    print(f"  Non-zero: {np.count_nonzero(npy_data)}/{npy_data.size}")
+    print(f"  Mean: {npy_data.mean():.4f}")
+    print("---")
     decodec_rlt=[]
     with torch.no_grad():
         decoded_waveform = codec_model.decode(torch.as_tensor(codec_result.astype(np.int16), dtype=torch.long).unsqueeze(0).permute(1, 0, 2).to(device))
@@ -457,6 +463,19 @@ os.makedirs(vocoder_stems_dir, exist_ok=True)
 vocal_output = None
 instrumental_output = None
 for npy in stage2_result:
+    # Debug: Check .npy file content
+    npy_data = np.load(npy)
+    print(f"Processing {npy}")
+    print(f"  Shape: {npy_data.shape}")
+    print(f"  Data type: {npy_data.dtype}")
+    print(f"  Min value: {npy_data.min()}")
+    print(f"  Max value: {npy_data.max()}")
+    print(f"  Mean value: {npy_data.mean():.4f}")
+    print(f"  Std value: {npy_data.std():.4f}")
+    print(f"  Non-zero elements: {np.count_nonzero(npy_data)}/{npy_data.size}")
+    print(f"  First few values: {npy_data.flatten()[:10]}")
+    print("---")
+
     if '_itrack' in npy:
         # Process instrumental
         instrumental_output = process_audio(