Code

audio_index.py

@@ -42,7 +42,7 @@ def get_embedding(row: dict) -> torch.Tensor | None:
         input_signal=audio_signal, input_signal_length=audio_signal_len
     )
     del audio_signal, audio_signal_len, audio_resampled  # Clean up resampled audio
-    return emb.detach().cpu()  # Return the tensor
+    return emb.detach().cpu().numpy()  # Return the tensor
 
 
 def get_embedding_from_array(sample_rate: int, audio_array: np.ndarray):